Energy Optimization of Radio - Lu

Energy Optimization of Radio

NGR Micro G1 B2/25

Authors: Kasper Ornstein Mecklenburg &

Anton Blomgren

2016

Master´s Thesis

Electrical Measurements

Faculty of Engineering LTH

Department of Biomedical Engineering

Supervisor: Jonas Bengtsson & Johan Nilsson

Company: Ericsson

Energy Optimization of NGR Micro G1 B2/B25

Anton Blomgren &Kasper Ornstein Mecklenburg

June 2016

1

Abstract

Rigorous studies have been conducted on Ericsson’s NGRMicro G1 B2/B25 radio. The radio’s part in the telecom-munications chain and components within the radio havebeen determined and mapped according to certain char-acteristics giving an understanding of their function. Theobtained knowledge has been applied in order to make theoperation of the radio more energy efficient. This has beendone by implementing new sleep modes and improving al-ready existing power saving features. The new sleep modeCell sleep, has taken the radio’s ”off state” from 32W to18W. The features that have undergone revisions and havebeen improved are TX micro sleep and MIMO sleep. Inshort the features scale capacity depending on demand.The original implementation of TX micro sleep consumed44W with some additions to the feature this has been re-duced to 38W and MIMO sleep has been reduced from 42Wto 39W. Combining the features gave a saving of 6W, from38W to 32W.

2

Acknowledgements

During the process of this thesis we had to acquire knowledge within many differentfields and this was possible due to the people working at Ericsson, in Lund but also inLindholmen and Kista. We want to thank our supervisor Jonas Bengtsson for alwaystaking his time helping us and discussing various subjects, Per Sanderup for repairingthe radio when we broke it and for sharing his expertise in hardware, Ulf Morland forhelp with SPI and clock measurements, Peter Nessrup for installing and showing howthe DU and radio software works, Robert Marklund for setting up an collaborative Latexenvironment on local servers, Henrik Sundelin for tips on how to solder properly, HansAndersson for supplying us with radios, the staff on 4:4 for being helpful and making usfeel welcomed and finally the personnel at Lund Institute of Technology.

Anton Blomgren & Kasper Ornstein Mecklenburg

3

Contents

1 Introduction 6

2 Background theory 82.1 Telecommunications and 4G/LTE . . . . . . . . . . . . . . . . . . . . . . 8

2.1.1 Sites and traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.2 FDD and TDD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.3 Data transfer and bandwidth . . . . . . . . . . . . . . . . . . . . 10

2.2 NGR Micro G1 B2/B25 . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.2.1 Downlink chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.2.2 Uplink chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3 Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.4 DC/DC converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.4.1 Measurement theory . . . . . . . . . . . . . . . . . . . . . . . . . 142.5 Current power save methods . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.5.1 Blocked cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.5.2 TX micro sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.5.3 MIMO sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Mapping the radio 16

4 Methodology 174.1 Equipment and lab setup . . . . . . . . . . . . . . . . . . . . . . . . . . 17

4.1.1 Radio NGR Micro G1 B2/B25 . . . . . . . . . . . . . . . . . . . 174.1.2 Measuring equipment . . . . . . . . . . . . . . . . . . . . . . . . 184.1.3 UE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194.1.4 Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.2 Minimum power consumption . . . . . . . . . . . . . . . . . . . . . . . . 204.3 Buck dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204.4 Potentiometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224.5 Possible applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.5.1 Cell sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234.5.2 TX micro sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234.5.3 MIMO sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5 Results 255.1 Current radio energy consumption . . . . . . . . . . . . . . . . . . . . . 25

5.1.1 Cell sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285.1.2 TX micro sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295.1.3 MIMO sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335.1.4 Feature combination . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.2 Buck dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

6 Discussion and conclusion 406.1 Buck dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406.2 Cell sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416.3 TX micro sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426.4 MIMO sleep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436.5 Combined features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436.6 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446.7 Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446.8 Making the radio intelligent . . . . . . . . . . . . . . . . . . . . . . . . . 45

4

7 Further work 46

8 References 46

9 Appendix 489.1 Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5

1 Introduction

Scientists and researchers have proved that the modern way of life has many negativeeffects on the climate and the environment. All over the world governments are adheringto the facts and strive to reduce the impact that our way of life has on the planet. Theseleaders are trying to establish national and international legislature in order to help thetransition to a more sustainable society.

In Sweden, for instance, the Ministry of the Environment has set up a number of goalsto reach by year 2020. The Ministry aims to have at least 50% of the consumed elec-tricity produced from renewable energy sources and energy efficiency should increaseby at least 20%. The European Union has reached an agreement and set up goals toincrease energy efficiency by 20% by 2020 and to have 20% of the electricity originatingfrom renewable sources [1]. A part from governments, international organizations likethe UN and Greenpeace push for the establishment legislature and agreements in orderto hasten the transition towards a sustainable future. There is an increased pressureon companies and institutions to adapt and become more sustainable, not only fromgovernments and organizations but also from the general public, as many become moreaware.

Many companies are seeing the benefits of a more sustainable way of business whichleads research and development towards new ideas and solutions. To optimize the use ofenergy and utilization of resources is economically beneficial as well as making productsmore attractive on the market. Some of the operators in Sweden claim to do their bestto minimize their impact on the environment and they have formulated goals of theirown. Tele2 for instance claim to make ”continuous efforts on reducing the energy con-sumption of the communication network” [2]. TeliaSonera has more concrete goals andaim to reduce their energy consumption with 10% per subscription equivalent, meaningthat all their different services should use 10% less energy [3]. In Germany operators aretrying to become more sustainable by reducing their green house gas emissions throughenergy optimization [4][5]. In Italy the same strive is proclaimed by some of the biggesttelecommunication operators [6][7][8]. All over the world there is an increased focus oncost and energy consumption from operators. [9].

Ericsson is competing on the global market and today 40% of the worlds mobile trafficpasses through their networks [10]. They sell their services to operators around theworld and in order to keep their grasp on the market their products need to be verycompetitive. There are many ways of having a competitive product for example; it canbe of the highest quality, have the best features or be the cheapest. With the pressureof a more sustainable society from governments, legislation and the public, another as-pect becomes increasingly important; the energy consumption of the product. With therapid technological evolution of modern society, a constant search for new products andimprovements has to be conducted in order for companies to stay in the forefront of themarket. This master thesis is such a search.

The master thesis strives to optimize and reduce the power consumption of Ericsson’sradio NGR Micro G1 B2/B25. The radio together with a digital unit (DU) make upthe link between the users mobile phone or user equipment (UE) and the core network.It is the link in a long chain which enables the connection to the World Wide Web. Itis also the part of the chain where most of the power is consumed, as can be seen inFigure 1 80% of the energy in the chain is consumed in the radio access network (RAN)[9].

6

Figure 1: Power consumption in the chain.

The RAN has the most opportunities of improvement, thus this thesis is focused onthe radio. The study is conducted on one of Ericsson’s smaller radios due to practicalreasons, a smaller radio has a lower output power and is therefore easier to handle. Lesscaution needs to be taken and the output signal requires less attenuation. Ericsson’sdifferent radio models are quite similar and share many components and therefore op-timization of the NGR Micro G1 B2/B25 is likely to be applicable on other models aswell. At the facility in Lund there is an advanced lab where simulations and real trafficscenarios can be run. A lab environment very similar to the environment in real liveRAN was setup and traffic scenarios with actual UEs were run.

The research has been an iterative process and in this report the result of the firstiteration is presented as background theory. During the first iteration an understand-ing of the radio was acquired and the components were mapped and categorized. Thistask has made up most of the research process and was extensive. Upon this ideas weredeveloped, applied and tested which make up the second iteration of the research process.

The report is divided into six sections. It begins with the background theory on whichthe rest of the report is based and this section is vital for comprehension. Then the labsetup and instruments used during the research are presented as well as our ideas ofimprovement. This is followed by a presentation of the results of these ideas and thenconcluded in discussion and conclusion. Subjects and ideas that were touched but notfinished are found under further work. At the end of the report the references are listedas well as an appendix with necessary abbreviations.

7

2 Background theory

The purpose of this section is to supply the reader with all necessary information abouttelecommunications with emphasis on long time evolution (LTE), how wireless data issent, theory on how to measure power correctly and methods Ericsson currently areusing to reduce power consumption in their radios.

2.1 Telecommunications and 4G/LTE

Ericsson’s LTE networks starts at the internet and ends with the wireless access pointcreated by the radio. It can be divided into four main parts:

• The core network which makes up the infrastructure of the network

• DU which controls the RAN

• Radios which transceives data

• UE enabling connection to the network.

The core network transports data between the servers of internet and out to the RAN.The RAN made up of many cells where each cell consists of at least one radio and oneDU. There are two main categories of cells with different functions; the coverage cellconsists of a powerful radio which can cover large areas, the capacity cell which has asmaller radio and covers a smaller area. The coverage cell supplies reception to a widearea and hands over traffic to the capacity cell which is placed strategically at high trafficareas to unload the coverage cell. The smaller cell is also used to supply reception toareas where the larger coverage cell can not reach, for example in a subway station. InFigure 2 the coverage cell can be seen covering a large area, the two smaller capacitycells are seen covering smaller areas; a subway station and a square.

Figure 2: Example of application of the different cells.

8

The DU has the intelligence in the RAN and controls the radios, the communicationwith the UEs and handles handovers between cells. It also handles all the adminis-tration of data, analysis and decision making, as well as interpreting data and handleserror control. It schedules the communication and decides which UEs talks and listensto which frequency and when. The radio simply converts the digital data from the DUto an analog signal and transmits it to the UEs and does the opposite in the otherdirection. The UE allows the end user to interact with the network and connect to theinternet oblivious of the advanced technology behind it. In Figure 3 the chain from UEto the world wide web can be seen, it also shows the communication from the UE tothe radio is refered to as uplink (UL) and from the radio to the UE as downlink (DL). [11]

Figure 3: Example of application of the different cells.

According to the LTE standard established by 3rd Generation Partnership Project(3GPP) each cell has to send Cell-specific Reference signal (CRS) with each DL sub-frame. There are a number of other reference signals which has to be sent at certainintervals in order to comply with the LTE 3GPP standard. This makes LTE quitetalkative because the radio sends out reference signals. [11]

2.1.1 Sites and traffic

The traffic going through different sites depends on their geographical location and alsowhat time of the day it is. During rush hour in a subway station the demand for capacityis higher than at night. Figure 4 gives an overview how much traffic goes through sites.The high traffic sites make out 10% of the sites and take 25% of all the traffic, 40% aremedium sites taking 60% of the traffic and finally 50% of all sites are low traffic site onlytaking 15% of the traffic. The high traffic sites have a demand of at least 25% capacity,the medium less than 25% and the low traffic sites down to no demand on capacity.Half of the sites have long periods of just idling and waiting for users to make use of thecapacity available.

2.1.2 FDD and TDD

There are two different ways of communication in LTE; frequency division duplex (FDD)and time division duplex (TDD) [12]. The principle for TDD (see Figure 5) is thatthe radio and the UEs communicate on the same frequency but never at the sametime. In countries where bandwidth and frequencies are sparse TDD is more common.TDD requires very high precision in time or else sending and receiving will overlap

9

Figure 4: How traffic is distributed on Ericsson’s sites.

creating noise and disturbance in the system. FDD (see Figure 6) is less prone todisturbances as it sends and receives simultaneously due to that the radio and the UEsalways communicate on different frequencies. It is more expensive in bandwidth due tothat the bandwidth is split between sending and receiving. [11]

time

fdl,ul

fre

qu

en

cy

TDD principle

RX TX RX TX RX TX

Figure 5: Principle plot for TDD.

2.1.3 Data transfer and bandwidth

The transmission time interval (TTI) for LTE is 1 ms and every TTI consists of 14symbols, i.e. a symbol is 1/14 ms or also 71.5 µs. During each symbol the radio has thepossibility to either transmit data or stay quiet. The amount of data sent during eachsymbol depends on three parameters; load, modulation and available bandwidth.

Depending on the quality of the signal to the UE and radio different modulation canbe used. Basically, if there is good reception it allows for a high quality signal and thetransmitted data rate is high and vice versa. So when the signal is poor or when theradio is signaling, QPSK is used and less data is transmitted. Higher modulated signalsare more sensitive to noise and disturbances. In Table 1 different modulations with

10

time

fdl

ful

fre

qu

en

cy

FDD principle

TX

RX

Figure 6: Principle plot for FDD.

related bits can be seen.

Table 1: Modulations methods with related number of bits.

Modulation Bits per symbolBPSK 1QPSK 216QAM 464QAM 6256QAM 8

The general term for packaging data higher than one bit is in-phase/quadrature (I/Q)data [13]. The principle is to use amplitude and phase to encode data onto a sine waveas

A cos(2πfct+ φ) = A cos(2πfct) cos(φ) −A sin(2πfct) sin(φ)

where fc is the carrier frequency, t is time, φ is the phase and also

I = A cos(φ), Q = A sin(φ)

A reference sine wave of the same frequency is necessary in order to identify the phaseand amplitude of the modulated signal. Once I and Q have been determined they willcorrespond to a certain coordinate in the complex plane. An example of a 16QAM 4bitmodulation scheme can be seen in Figure 7.

Apart from the modulation used, the bandwidth determines the data transfer rate.LTE is defined for bandwidths of 1.4-20 MHz and the spacing between sub-carriers inthe bandwidth is 15 kHz [12]. A single sub-carrier during a symbol is referred to as aresource element and an UE is assigned a certain number of elements creating a resourceblock. A general formula to calculate the data transfer rate is defined as the numberof symbols per second times the bandwidth divided by 15kHz and finally multiplied by

11

I, real axis

Q,

ima

gin

ary

axis

16QAM

Figure 7: 16QAM 4 bit modulation scheme.

the bits which depend on the type of modulation used. A 20MHz bandwidth LTE radiousing 16QAM gives

14 · 1000 · 20 · 106

15 · 103· 4 ≈ 74.7Mbit/s. (1)

An important note is that this is the theoretical calculated value where all resourceelements are allocated to data transfer, however in practise some resource elements areused for control and signaling to ensure that correct data has been received [14].

2.2 NGR Micro G1 B2/B25

The radio operates on band 2/25, which are two bands at similar frequencies used inthe USA. The frequencies span from 1930Mhz to 1995Mhz for downlink and 1850Mhzto 1915MHz for uplink [15]. The Micro has a bandwidth of 20MHz and is capable of64QAM modulation in downlink giving a maximum throughput of 150Mbit/s and inuplink the bandwidth is 10Mhz and with a maximum of 16QAM modulation it has50Mbit/s throughput. The radio uses a cavity filter which only lets desired frequen-cies pass through (bandpass filter) on both up- and downlink. It is connected to a DUthrough an optic cable and communication is done through common radio protocol in-terface (CPRI).

The radio is designed to be used in a small cell, therefore it is smaller and has loweroutput power level compared to the Macro radio which is used as a coverage cell. Theradio has two antennas and two sets of TX and RX chains (the RX chains share re-sources) in order to enable multiple output multiple input (MIMO) which allows for ahigher throughput. The output power of the Micro is 5W on each of the antennas, theMacro has typically an output power of 40W per antenna. The two different radios havesimilar hardware and some components are identical even though they have differentareas of application. Below follows a synoptic explanation of the downlink and uplinkchains inside the radio.

12

2.2.1 Downlink chain

The DU receives a data package from the core network. It decodes the recipient andencodes the data to the same frequency as allocated to the UE, the packet is then sentthrough the CPRI link to the radio. The radio receives the package and converts itfrom a digital signal to an analog. It also converts the signal to the higher frequencyband of the radio, in other words it lifts the signal up to the carrier frequency. Thesignal is then amplified in the power amplifier (PA) and in order to have the outputpower specified by the DU, the signal from the PA is fed back for control to ensure thatthe correct power level is met. The signal is now transmitted to the UE through oneof the antennas. The UE receives the package and if the error control is positive noretransmission is necessary, otherwise the DU will resend the data and possibly changeto a lower modulation. [16]

2.2.2 Uplink chain

The UE sends a data package on a specific frequency assigned by the DU. The radioreceives the signal which is amplified in two stages before it is downconverted. In thedownconversion the carrier frequency is removed, the signal is converted down to lowermore manageable frequencies called intermediate frequency (IF). The signal is thenpassed on to a analog to digital converter (ADC). The signal is now digital and in asuitable format for the DU to which it is passed on for analysis and processing. Ifnecessary the DU will ask the UE to resend the data with an increased signal strength.If the data is intelligible it is passed on to the core network and on to its destination.[17]

2.3 Clocks

A clock is an oscillating circuit supplying an AC signal at a certain frequency. Theclocks drive the digital circuit, it is the pulse of the system. It supplies a mean for syn-chronization and drives the operation of the components. A clock can generate a signalby sending a current through a crystal which then starts to vibrate and this vibrationdecides the frequency of the clock signal. Some oscillators can vary the frequency of thegenerated signal one example is the voltage controlled oscillator (VCO), where the inputvoltage can adjust the generated frequency. These systems are often combined with aphase locked loop (PLL) to synchronize the generated clock with a reference clock andthereby the rest of the system. The PLL supplies feedback to the VCO which regulatesthe clock frequency until the PLL has locked on to the reference clock. When a lock hasbeen established the clock signal is stable and in sync with the rest of the system. VCOsare often used in components that need a higher frequency than the rest of the system,if a lower frequency is needed instead a divider can be used. The divider reduces thefrequency by the integer chosen to divide it with. [18]

2.4 DC/DC converters

Buck converters are often used as an efficient mean of creating lower power domainsfrom higher voltage supplies. The converter is an analog device which rapidly switcheson and off, commonly a MOSFET transistor which creates a square wave. In orderto supply a steady voltage the square wave charges an inductor and capacitor whenhigh and when the square wave is low the inductor and capacitor discharge, creating asmoother voltage. This creates ripple in the output voltage which is reduced by couplingcapacitors between supply and ground. Buck converters can be very efficient and some

13

achieve efficiencies of up to 95-96% [19]. The efficiency is dependent on a few fixed pa-rameters and therefore they are designed to be used under specific conditions [20]. Oncethe specific conditions have been set the efficiency curve depends on the current pass-ing through the converter. One must therefore approximately know the power passingthrough a power domain in order to choose appropriate parameters for a buck converter.

The buck converter has a small DC ripple which is hard to remove entirely, so when avery consistent voltage supply is required a low-dropout (LDO) regulator is used instead.The LDO regulator is a linear voltage regulator which supplies a very consistent andripple free voltage and is commonly used to supply for example VCOs. The efficiency ofa LDO is directly proportional to the input output ratio. If the input is 5V and outputis 3V the efficiency is 60%; a very wasteful converter compared to the buck. [21]

2.4.1 Measurement theory

To determine the current flowing through a circuit, the voltage drop over a resistor withknown resistance is measured. If the voltage has no reference to ground it is called adifferential measurement and using Ohm’s law the current is calculated according to

I = ∆U/R (2)

where I is the current in A, ∆U the voltage drop in V and R the resistance in Ω. For thisequation to accurate results it is important that the measurement is done properly. InFigure 8 the solderings avoid the contact resistance which is present between the PCBand the resistor. This way of measuring the current is called a 4-pole measurement andavoids the involvement of the contact resistance. This is due to that no current will passthrough the voltmeter and therefore the effects of the solderings are eliminated. Thepower is finally calculated according to

P = U · I. (3)

Figure 8: The grey area is the contact resistance between PCB and resistor, and theblack where the solders should placed to avoid measuring the contact resistance.

14

2.5 Current power save methods

There are some measures of power save functionality already available for the Micro andother radios. Below follows a short description of the features that are implementedtoday.

2.5.1 Blocked cell

When there is no traffic on a radio and it is known approximately when there will beagain the radio can be blocked. In this mode of operation the radio neither transmitsnor receives and it is for all intents and purposes switched off. Since it is known whenthe radio needs to be up again boot time is no concern. [22]

2.5.2 TX micro sleep

The radio is not always transmitting, depending on the traffic and the scheduling ofthe transmissions there will be windows in time when the radio is quiet. During thesewindows the biasing to the final stage of the amplification is turned off. This is doneby generating a strobe signal based on information about the data to be sent from theDU. The information is sent in a message which specifies at what symbols the radio isto transmit during that TTI. This strobe signal toggles the biasing, turning it on andoff. [23]

2.5.3 MIMO sleep

When the traffic is low the full capacity of the radio is not needed. This allows for oneof the TX chains to be switched off. The throughput of the radio is thereby halved.[24]

15

3 Mapping the radio

This part of the report have been removed due to confidentiality of Ericsson products.In this part the inner workings of the radio is presented and this process was very timeconsuming and made up a large portion of the research process.

16

4 Methodology

After going through the radio and its components in the first iteration, findings wereanalyzed and possible applications of these findings were discussed and implemented.In order to test and evaluate these findings a laboratory environment was setup. Thesetup allowed for real traffic scenarios to be run on the actual equipment which makesup the RAN in live networks. Below the lab setup and equipment is presented and afterthat the possible applications established for the second iteration are introduced.

4.1 Equipment and lab setup

XYZ is a confidential component in the radio and its name has been changed.

4.1.1 Radio NGR Micro G1 B2/B25

In the lab setup a NGR Micro G1 for band 2/25 of version P1C running software withPid CXP9013268%9 R62SB01 has been used. It has been supplied with power from anexternal source instead of the original power supply to give exact measurements of theconsumption of the radio without the losses of the initial power transformation. Theradio is open to enable direct access to the PCB. The PCB is attached to the bottom halfof the chassis which has a heatsink and the chassis is placed directly on the table. All ofthe shields has been removed in order to be able to access components. It is cooled byan external fan since the power consumption of the fan unit is not relevant for this study.

The opened radio is connected to its cavity filter through RF cables as shown in Figure9. The filter in turn, is connected to a series of attenuators. Each branch is connectedto 20dB attenuators with a max effect of 50W. The radio’s peak effect makes it nec-essary to have an attenuator of that power level first in the series to avoid damagingthe equipment. Branch A is then connected to three smaller attenuators in series in thefollowing order: 10dB, 30dB and 10dB. Branch B was attenuated with two 20dB atten-uators in series. It was unintentionally attenuated 10dB less however this did not affectthe results. The two RF cables then connect to splitters with another 10dB attenuation.The splitters split the signals in to four which then is connected to four UEs. Figure 10shows the lab setup and it is presented schematically in Figure 11. Note that the DU isaccessible through the local network.In order to make detailed measurements a number of wires were soldered to certaincomponents and on to splines with pins glued to the side of the bottom half of the chas-sis. In Table 2 the resistors connected to the spline with pins for detailed measurementof the DC/DC block are listed together with their power domain. The resistors werechosen because they were easy to access and on the different power domains’ path.

Two wires were soldered on to each of the resistors, one before and one after the resistorto allow for a differential measurement. This is all done in accordance to the 4 polemeasurement method. Connections were also made to certain signals of interest.The radio is supplied by 36V through a modified RPM 777454/0200 cable. The cableconnector that connects to the original power supply (PSU) has been replaced by twobanana connectors.

For direct communication a testcard controlled through a program called Term-R4B25was used. Moshell, an Ericsson internal command tool, is used to control a DU 4101which in turn controlled the radio through the CPRI link. The CPRI link is connectedinto the SFP A slot.

17

Figure 9: The opened radio connected to the cavity filter.

Figure 10: Lab setup.

Table 2: The resistors and their resistance, named as in the schematics [25].

Resistor Resistance, mΩ Power domain, VR337A10 2.5 5.1R1003A10 5 Switch 5.1R537A10 5 3.7R534A10 2.5 3.3R1215A10 15 3.0R536A10 5 1.8R643A10 5 1.35R645A10 5 0.982R646A10 2.5 0.768-0.877R435A10 15 16-32

4.1.2 Measuring equipment

The PSU used was an Agilent Technologies N6705B with dual power cells to supply 36Vand up to 3.5A and one of the other channels was used for measuring differential voltage

18

Figure 11: Schematics of the lab setup.

over the PA buck converter. For the measurements of the remaining buck converters aNational Instruments PXIe-1071 with two modules each with 8 differential measurementchannels was used. It has high sampling rates of up to 50MHz. The PXIe was controlledby a computer running Ericssons own developed analysis tool, a labview program calledPower Analyzer v0.29. The PXIe has a max limit of voltage input (around 10V) eventhough the voltage difference only is in order of mV over the resistors. Due to thisan Agilent 34401A multimeter was used for higher voltages, a Labview program waswritten to gather data and control the instrument. To study clocks and other signals aYokogawa DL9240 oscilloscope was used together with Techtronic probes 10X, 10MΩ,8.0pF and 500Mhz bandwidth.

4.1.3 UE

In order to simulate and test traffic on the radio, four UEs were used and can be seen inFigure 12. The UEs were connected to a PC which runs the program LINS3 used to setup the UEs. To simulate traffic data was sent and received from a server via Iperf3.0.11win64 or online speed tests and general web browsing. The UEs were directly connectedto the radio through RF cables with attenuators connected, the signal strength perceivedby the UEs was approximately -80dB. The connection of four UEs enabled a maximumload of the radio on both up- and downlink.

4.1.4 Control

Interaction to the radio was done through two different interfaces; the testcard withdebugging interface and the DU with the Moshell interface. Term-R4B25 was easy touse and it was possible to create macros containing multiple commands. The testcardalso eliminated the need of CPRI communication which allowed turning off componentsessential to the CPRI communication. The use of the DU with the Moshell interfaceallowed for remote operation otherwise the two enabled the same internal control of theradio. The features TX micro sleep and MIMO sleep were activated and deactivated viathe DU. The DU was running CXP102051/25 R7FJ with an added features enabling TX

19

Figure 12: Four UEs stacked on top of each other.

micro sleep and another enabling more advanced power measurements from the internalsupervision.

4.2 Minimum power consumption

After the first iteration the components’ function and how they were controlled hadbeen established. In order to get an idea of how much power could be saved by turningoff a component, putting it in standby or sleep mode the total power supplied to theradio was measured while components were turned off, one after another. The radiowas in Blocked cell mode when conducting this study, it would otherwise reboot whenits circuits went offline. This essentially created a new more effective Blocked cell whichwas named Cell sleep and represents the minimum power consumption of the radio whilestill being rebootable via the CPRI interface.

4.3 Buck dimensions

After learning that the DC/DC block is similar to the one as in the much more powerfulMacro radio the dimensioning of the buck converters came into question. The buck con-verters are designed to work in specific conditions in order to be as effective as possibleand as the smaller radio consumes less power the buck converters might be dimensionedincorrectly.

In order to justify the dimensions of the buck converters, a few tests were designedto check their efficiency. By measuring the current going through resistors in Table 2and the total current going in to the radio from the Agilent N6705B power supply anestimate could be acquired. To measure the current passing through the resistor, andthereby the buck converter, the voltage before and after the resistor and its resistanceneeds to be known. The first 16 channels of the PXIe were then connected to the splineson the side of the radio. The buck converter supplying the PA was connected to theAgilent N6705B power supply channel B. The PA buck output voltage is adjustable and

20

is adjusted by the iWarp so in order to be able to conduct the calculations the voltagesource needs to be known, i.e. the voltage coming out of the converter. The voltage wastherefore measured by the Agilent 34401A multimeter.

To study the conditions the radio’s converters were working under a set of traffic loadscenarios were designed, they are presented in Table 3.

Table 3: Test traffic scenarios.

Name Total loadFull 100%Half 50%Quarter 25%Ten 10%Idle 2%Blocked cell 0%Cell sleep 0%

To create the different scenarios four UE’s were used to send and receive data from aserver via Iperf, two sending and two receiving. The different loads were created bytaking the theoretical max throughput on both up- and downlink (50 and 150Mbit/s)and dividing it with 2, 4, and 10. This gives the throughput for half, quarter and tenpercent load. Iperf was then configured to try to send and receive at those throughputsfor a certain amount of time. The Idle scenario was created by simply letting the UE’sidle, no added traffic from Iperf, however the PC might have sent and received smallamounts of data. In the Blocked cell scenario the radio is not transmitting nor receivingand in the Cell sleep scenario the new Blocked cell implementation is run.

The radio was connected to a computer through the testcard. The PSU was connectedto the same computer and the official software was used to collect data from the powergoing in to the radio and the current through the PA buck. The sample rate of the PSUwas set to 1024 SPS. The multimeter was connected to the same computer and datawas setup to be collected at 3Hz. The sample rate of the PXIe was set to 10kHz andbefore booting the radio it was calibrated to eliminate DC offset and just before thefirst scenario it was calibrated in the regards of temperature. After booting the radiothe UEs was started and setup through LINS3 on the connected computer.

The traffic scenarios were run for 300 seconds from which 30 seconds of data was col-lected. Before the new scenario the PXIe was again calibrated in regards of temperature.Iperf was taking too long to connect to be able to run the scripted scenarios so they hadto be run manually.

The conversion is done in steps, where in the conversion steps the measurement pointsfor where the DC/DC units were made can be seen as circles in Figure 13. By measuringthe power going in to the radio and what comes out after each step of conversion theefficiency of each step can be estimated. The measurements were divided into threeaccording to the steps of conversion; the first the PSU, the second the PA and TRXbuck converters and the third the remaining buck converters and the 5.1V switch. So bydividing the power measured after step two (total from all converters) and dividing itwith the power going in to the radio the efficiency of step to can be calculated. The sameapproach gives the efficiency of step three but the power coming in to these converters

21

is the power coming out of the step two converters. The total efficiency was calculatedby dividing the power coming out of step three by the total power going in to the radio.Due to accessibility it was not possible to measure before and after every individual buckconverter. This means that only the efficiency for the entire step can be calculated, notfor individual converters. Figure 14 is a picture of a resistor seen through a microscopeand makes a 4-pole measurement possible.

Figure 13: Circles show where the measurements of the power domains were done.

Figure 14: Solderings on a resistor used for a differential measurement.

The collected data was then imported to MatLab to be processed and analyzed. A lowpass filter was added when necessary to eliminate disturbances and make the data easierto interpret.

4.4 Potentiometers

One idea was to continuously change the voltage to the two adjustable buck converters.In order to do this the rise and settling time of the potentiometers voltage needed tobe known. The measurement was done by using the same measurement points as forthe differential measurement (R646A10 and R435A10) for the buck converters with thedifference that the measurement had a ground reference.

4.5 Possible applications

With the knowledge of the first iteration at hand, a search for possible applications wasinitiated. A natural place to start was the already implemented features since it came to

22

be obvious that they could be improved. The features were studied more thoroughly andadditions to them was introduced and tested. A thorough presentation of the processfor the different features are presented below.

4.5.1 Cell sleep

The feature is only to be activated when the radio is not being used and it is knownwhen it again is needed. The idea was to simply reboot the radio at a preemptive timewith regards of the boot time. This means that the boot time of the radio is no concernand the goal became to turn off as many components as possible while retaining com-munication via CPRI.

The earlier findings of minimum power consumption were reworked and some addi-tions were made. The process was the same, turn off the component and study theeffect on the total power consumption, if CPRI communication is lost; take a step back.


In the current implementation only the biasing of the final amplification stage is turnedoff and it is very fast, it has a very short response time, which is needed since the short-est time windows are 71.5µs (one symbol). This puts demands on what can be includedin the feature, the switching on and off has to be fast and the component has to comeback up fast. The idea was to turn off as many components as possible which wherefast enough to be turned off and able to come back up again, all with in the time of asymbol. Different software and hardware solutions do to this needed to be found. Withthis in mind components in the TX chain were studied once again.

In Table 4 the relevant components and their theoretical response times are presented.The table covers the components which are physically possible to turn on and off insuch a short time period with out consideration of the consequences, such as stabiliza-tion times for PLLs.

Table 4: Candidates for TX micro sleep

Removed due to confidentiality.

The theoretical response times for the DAC and ADC are taken from their productspecifications and can be considered factual for optimal conditions [26][27]. The timesfor the clocks are estimations of how long it takes for a gated clock to come back upand judged to be a few clock cycles. No consideration to the resulting behavior of thecomponents losing the clock signal is taken. The PA driver and predriver are turned offby closing the biasing and thereby very fast, consequences on the output signal is notconsidered. The XYZ’s response time is dependent on how the pin used to power downis programmed.

The next step was to figure out where and how the new components could be added inthe source code. The response times of the components also needed to be determined.

4.5.3 MIMO sleep

When the traffic is below a predetermined threshold the radio can be put in MIMO sleepmode, this means that only one of the TX chains are used. Both RX chains remain op-erational since they are, as mentioned earlier, for all intents and purposes a single chain.

23

The threshold design enables longer response times since it allows for a preemptive startup of the closed down chain.

The process of designing an improved MIMO sleep was similar to that of the cell sleep butinstead of making sure CPRI communication was maintained the retention of through-put was the parameter differentiating success from failure. The four UEs were connectedand setup as in the earlier tests. Measurements of the power consumption was collectedfrom the PSU.

24

5 Results

In order to be able to compare results to each other the power consumption for the radiowith no feature, the original feature and the new feature will be presented. The valuespresented are with the testcard still in the radio, the testcard consumes 0.5W so theactual consumption is 0.5W lower than the measured value. If nothing else is stated thenumbers are mean values over a 30 second period.

UVW and PQR are confidential components in the radio and their names have beenchanged.

5.1 Current radio energy consumption

The radio’s power consumption depends on the traffic load and during testing it peakedat 81W at maximum load. The consumption of the radio and PA in the traffic scenariosis presented in Table 3 and shown in Figures 15 and 16 respectively.

0 20 40 60 80 100 120 140 160 180

time, s

30

40

50

60

70

80

po

we

r, W

PSU 36V

full half quarter ten idle blocked

Figure 15: The PSU output power in the different scenarios.

In Table 5 the power consumption for the different scenarios with no features is listedand also shown in Figure 16.

Table 5: The power consumption of the radio in the different scenarios.

Scenario Power consumption WFull 73.43Half 67.45Quarter 60.52Ten 56.12Idle 53.15Blocked cell 31.85

To show the difference between the Idle and Quarter scenario, heat pictures were takenwhile running. Figure 18 is in Idle and Figure 19 is in Quarter and shows increased heatin the PA and TX low blocks.

25

0 20 40 60 80 100 120 140 160 180

time, s

0

5

10

15

20

25

30

35

40

45

50

po

wer,

W

PA 22-29V

full half quarter ten idle blocked

Figure 16: The power consumed after the PA buck in the scenarios.

0 10 20 30 40 50 60 70 80 90 100

capacity, %

15

20

25

30

35

40

po

we

r, W

PA power

Figure 17: Power consumption plotted against load.

26

Figure 18: Heat picture of the radio in Idle mode.

Figure 19: Heat picture of the radio Quarter capacity mode.

27

5.1.1 Cell sleep

The components that were included in the final implementation and their power downmodes can be seen in Table 6.

Table 6: List of components and power down modes included in Cell sleep.

Confidential.

The clocks (clks) in Table 6 refers to all clocks on that specific branch. The power con-sumption in Cell sleep is listed in Table 7 together with the Ericsson implementation.As the table shows the ”off state” of the radio was reduced by 42.7%.

Table 7: The power consumption for the different Cell sleep implementations.

Implementation Power consumption WEricsson 31.85New 18.23

Heat pictures were taken to highlight the differences between the two sleep modes.Figure 20 is the Ericsson implementation and Figure 21 is the new Cell sleep implemen-tation. The pictures show the heat signatures of the components cooling and a weakersignature from the UVW and DC/DC block.

Figure 20: Ericsson Blocked cell implementation.

28

Figure 21: New Cell sleep implementation.


The first idea for the improvement of this feature was based on the theory that theTX micro sleep feature that is implemented was a snippet of code. This would allowfor a few simple additions of the candidate components to that code since all interruptroutines and such would already be handled. When studying this further it becameapparent that it was in fact a hardware implementation,. It required no lines of code,only a few register writes for initial setup was necessary. The radio’s FPGA was used togenerate a strobe signal which was the propagated SIG1 signal. The signal is generatedfrom the data sent in the SF-info message sent by the DU. However further researchshowed that there was hardware infrastructure to allow for a similar implementation ofthe predriver and driver. In order to understand and to be able to study the behaviourof the feature, measurement points were added to the SIG1, SIG2 and SIG3 signals.The strobe signal for SIG1 was then studied with the oscilloscope and can be seen inFigure 22.

In Figure 22 a single TTI for the strobed SIG1 signal is shown. The radio is currentlyin idle mode and when the signal is high the radio is not transmitting and when thesignal is low it is transmitting. As the radio is in idle mode the pattern seen is the CRSsignaling pattern. The FPGA was setup to propagate to the driver as well. For thepredriver the signal was inverted since it is turned on when the signal is high and turnedoff when it is low. In order to incorporate the other candidates presented in Table 4,software would have to be written.

To be able to include the gating of the DAC clock, stabilization time of the PLLs forthe PQR LO and XYZ A needs to be determined. In a discussion with Kent Persson,who works with designing the ASIC, he pointed out that the execution of software will

29

0 100 200 300 400 500 600 700 800 900 1000

time, µs

0

0.5

1

1.5

2

2.5

3

3.5

volta

ge

, V

Strobe pattern

Figure 22: The strobed SIG1 signal.

be too slow to be included in TX micro sleep. The execution and transmission of theSPI commands or GPIO toggling can not be guaranteed to finish in the specified timeintervals. Due to this all ideas of shutting off components controlled by GPIO or SPIwere abandoned.

The XYZ however is included in the TDD functionality and has a programmable pinPIN1. The TDD block generates a strobe signal as well although it works with longertime frames. The TDD switches at TTIs or milliseconds while the TX micro sleepswitches at symbols or microseconds. If the TDD strobe signal could be programmed toswitch at symbol level the XYZ could also included to the TX micro sleep. Unfortunatelythe hardware in the TDD block does not have the capability to generate a strobe signal.However, the two XYZs consume about 2.5W each, it would be beneficial to includethem in TX micro sleep. In order to test and demonstrate how beneficial this could be,two solderings to the PIN1 pins were made allowing for a hardware workaround. Thiswas done by disconnecting the original PIN1 paths from the ASIC by removing the re-sistors R6A2 and R6A3, and instead connecting the SIG2 strobe signal to the PIN1 pinto the XYZ, thus both were included in the TX micro sleep. In Table 8 the mean powerconsumption for the different version are presented and in Figure 23 the measured datais shown.

30

Table 8: The power consumption of the different implementations in Idle mode.*Due to the first radio breaking this value is measured using another radio.

Implementation Power consumption WFeature off 53.15Original 44.43Drivers 41.63*XYZs 39.44

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

time, ms

35

40

45

50

55

60

po

we

r, W

TX micro sleep

mean :44.43W

mean :39.44W

Figure 23: TX micro sleep in idle with mean values for original and new implementations.

The implementation ”Drivers” is the original implementation with the addition of thedrivers and predrivers. The implementation ”XYZs” is the ”Drivers” implementationwith the addition of both the XYZs. This results in the final implementation of TXmicro sleep.

To visualize the difference between the implementations heat pictures were taken. Fig-ure 24 is the original implementation to be contrasted with Figure 25 which is the newimplementation. The radio is in Idle mode when the pictures are taken and had reacheda stable temperature.

A test using the scenarios in Table 3 was also done and the throughput and powerconsumption was logged. The power consumption is presented in Figure 26 and thethroughputs were as expected, in other words exactly 50%, 25% and 10% of the radio’smaximum capacity. In Table 9 the values are presented with corresponding savings in%, note that the values in ”No feature” are the same as in Table 5.

Table 9: PSU power in W with no features compared to the improved TX micro sleep.

Implementation Half Quarter Ten IdleNo feature 67.5 60.6 56.1 53.1New 66.2 55.7 48.8 38.7

% saved 1.9 8.1 13.0 27.2

31

Figure 24: Ericsson TX micro sleep implementation.

Figure 25: New TX micro sleep implementation.

32

0 5 10 15 20 25 30 35 40 45 50

capacity, %

35

40

45

50

55

60

65

70

po

wer,

W

PSU power

Original

New

Figure 26: Original and new PSU power at different capacities.

5.1.3 MIMO sleep

The components that were included in the final implementation is presented in Table10.

Table 10: Components that are turned off in MIMO sleep.

Confidential.

Table 11 shows the mean power consumption of the original implementation and theimproved one. These values are the mean calculated from data which of some can beseen in Figure 27.

Table 11: The power consumption of the MIMO sleep implementations.

Implementation Power consumption, WEricsson 42.38New 39.43

The radio was in Idle mode. Figure 28 shows the heat signature of the original imple-mentation, Figure 29 shows the new version.

33

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

time, ms

38

39

40

41

42

43

44

45

46

47

po

we

r, W

MIMO sleep

mean :42.38W

mean :39.43W

Figure 27: MIMO sleep data.

Figure 28: Original MIMO sleep implementation.

34

Figure 29: New MIMO sleep implementation.

5.1.4 Feature combination

The biggest improvement is achieved when combining TX micro sleep and MIMO sleep.The effects of TX micro sleep is reduced since there only being one active branch whenMIMO sleep is activated. However the consequences of closing down one branch letsthe combination come down to consumption levels close to that of the Ericsson im-plementation Blocked cell while still being operational. Table 12 presents the originalimplementations combined, the new and the Ericsson Blocked cell. The measurementdata for the original and the new implementations is shown in Figure 30.

Table 12: The power consumption of the different feature implementations combinedand the Ericsson Blocked Cell for comparison.

Implementation Power consumption WEricsson 38.08New 32.61Blocked cell 31.85

This once again highlighted with heat pictures, Figure 31 is the combination of theoriginal implementations, Figure 32 is the combination of the new.

35

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

time, ms

30

32

34

36

38

40

42

44

46

po

wer,

W

TX micro sleep and MIMO sleep

mean :38.08W

mean :32.61W

Figure 30: TX micro sleep and MIMO sleep with mean values for the original and thenew implementations.

Figure 31: Combination of Ericssons implementation of TX micro sleep and MIMOSleep

36

Figure 32: Combination of the new implementations

5.2 Buck dimensions

The PSU supplies both the PA and TRX buck converters and measurements are doneat the output of the converters, see Figure 13, due to this the power of these two buckconverters are added together when calculating the efficiency. The PA and TRX buckconverters’ power and efficiency can be seen in Table 13. The total is calculated byadding PA and TRX together and the efficiency by dividing this total by the PSUmeasured power.

Table 13: PA and TRX buck power in W and the efficiencies in % at different modes.

Confidential.

The power going through the TRX buck is more or less consistent during different modesuntil the radio is set in blocked cell. This can be read out from Table 13 and seen inFigure 33. The data for full, half, quarter and ten will not be presented for the remain-ing buck converters as they show the same values in these modes as in Idle. In Table14 the combined efficiency is calculated for the remaining buck converters. The powergoing through the buck converters were added to create the total which is divided bythe power passing through the TRX buck converter. An important note is that thepower passing through the switch is subtracted from the TRX and not added to theother buck converters when calculating the efficiency since it is not a converter. Themean values in Table 14 are calculated from the data shown in Figures 34, 35, 36 and 37.

The potentiometers that adjust the voltage for the two adjustable buck converters weremeasured and their settling and rise time can be seen in Figures 38 and 39. The value

37

Figure 33: Confidential.

Table 14: Buck power and efficiencies at different modes.

Confidential.

(a) Confidential. (b) Confidential.








written to the potentiometers was 0 to 255 and 255 to 0, i.e. the maximum possiblerange.

38

0 0.5 1 1.5 2 2.5 3 3.5 4

time, ms

0.75

0.8

0.85

0.9

0.95

vo

lta

ge

, V

VCC settle/rise time for voltage

0 0.5 1 1.5 2 2.5 3 3.5 4

time, ms

0.75

0.8

0.85

0.9

vo

lta

ge,

V

Figure 38: Rise and settling time for the potentiometer.

0 2 4 6 8 10 12 14 16 18 20

time, ms

15

20

25

30

vo

lta

ge

, V

PA settle/rise time for voltage

0 2 4 6 8 10 12 14 16 18 20

time, ms

15

20

25

30

35

vo

lta

ge

, V

Figure 39: Rise and settling time for the potentiometer.

39

6 Discussion and conclusion

In the conducted study limits were pushed and some possible issues neglected. Thisapproach was necessary in order to be able to find new possibilities and change thingsthat always have been in a certain way. During the process we encountered setbacksripping ideas apart however by adapting these ideas some could still be implemented.The ideas that were not possible to implement are discussed and noted, and hopefullyfuture hardware and software designers will take them into account when designing nextgeneration radios.

We have not taken RF signal quality and wear and tear of components into accountin the study. When checking whether or not a feature worked we tested if we still couldcommunicate with the radio and if we had the desired throughput. Basically we createdproof of concepts and were satisfied if the radio functioned while running the feature.

The trial and error method that was used might have come with some unwanted andundocumented consequences. When using the ”I wonder what this does?” approach andthe only parameter for evaluation is the power consumption you might end up with aradio in a quite undefinable state. Some registers might not be reset at a reboot andtherefore contain values they are not expected to contain. When writing to registersand sending SPI commands the wrong address or values might have been used giving anunexpected and possibly undocumented results. The original radio finally gave up on usand we decided to modify a new radio to allow for further demonstrations. We are notsure why the radio stopped working but one thought is that it could be the solderingsdamaged the PCB. The damaged did not occur straight away but over time becausethe radio would act normally and then during operation it would restart or not givemaximum throughput. The new radio differed in power consumption in the differenttests run compared the first radio. In general it would consume slightly less, the reasonfor this is unknown.

6.1 Buck dimensions

During operation the DC/DC buck converters are very efficient with efficiencies rangingfrom 93-95%. At first we thought that these values were incorrect because we expectedthat the buck converters were dimesioned for the larger Macro radio. The results how-ever came out the same after multiple measurements. The resistors have a 1% marginof error so if all the buck converters in step three (see Figure 13) would have been 1%below the specified value the efficiency would only have changed 1-2%. This scenario isvery unlikely so instead we need to look at the buck converter’s properties. In Figure40 an efficiency plot for the 1.8V buck converter is shown. The efficiency is higher if thedifference between input and output voltage is smaller and is almost consistent between1 and 3 A, looking at Vin = 5V. It is only when the current is less than 0.5-0.6A thatthe efficiency decreases below 90%.

In Table 14 the power per power domain is listed. In order to follow the efficiency curvethe power is converted to current, as seen in Table 15. An important note is that Figure40 looks different for all the other buck converters. As the power decreases from e.g. 3Ato 2A the efficiency will increase, so only once the current is below a certain level theefficiency will quickly decrease. This concludes that there is definitely a possibility thatthese high efficiencies are correct.

We had ideas to adjust the two adjustable buck converters voltages however as seen in

40

Figure 40: Efficiency plot for the 1.8V buck converter [19].

Table 15: Current through the buck converters in sleep mode.

Confidential.

Figures 38 and 39 it takes very long time for the voltage to reach the desired value. Theapplication areas for these adjustments are therefore limited.

6.2 Cell sleep

In Table 14 the power in Cell sleep is 14.7W for step 3, in Table 13 15.47W for step 2 and18.24W for step 1. The steps can be seen in Figure 13. The losses in step 1 to 2 and 2to 3 alone are around 20% or 85% and 95% respectively. The buck converter’s efficiencydepends on the current, which decreases as more components are turned off, motivatinga low-current buck converter as the radio enters cell sleep. So a solution to this wouldbe to have two 5.1V converters, one which operates at normal use (high currents) andone at cell sleep (low currents). This would increase the DC/DC unit efficiency in cellsleep and the radio would consume less energy. For example the efficiency of the TRXbuck converter would be improved by 5% (from 85% to 90%) in Cell sleep, the overallefficiency would go from 80.8% to 85.5% and we would save an additional 1W, droppingdown to 17.2W from 18.24W.

The optimal power save feature for cell sleep would be a ”wake on CPRI” function,i.e. the radio is entirely shut off with the exception of the SFP which maintains con-nection with the DU. The SFP should be able to boot the radio when told to do so,this would require quite large changes in the hardware and software. The radios todayare designed assuming the unit is constantly turned on generating heat and keepingmoisture out, so condensation is an issue that will have to be addressed in new a design.By having a separate DC/DC branch only powering the SFPs and utilizing the alreadyexistent DC/DC supervision circuit the feature could easily be implemented. When theradio is needed a signal would be sent through the CPRI communication, the SFP would

41

respond by sending a signal to the DC/DC supervisor which powers up the remainingDC/DC units, booting up the radio as normal.

6.3 TX micro sleep

After we had completed cell sleep we had acquired enough knowledge to create a powersave mode in which the radio runs and functions. After hearing about TX micro sleepwe started looking into which components that could be shut down in the small windowof a symbol. We understood there were quite many possibilities as most componentshave very short response times. However when we started digging deeper we realizedthat there were some limitations that would difficult to overcome.

A lot of time was spent trying to measure the lock time for PLLs and in the endwe did not find any satisfying method to do so. The clock signal going into the clockbuffer from the PQR and the output LO signal from the XYZ (this clock locks to thesignal from the clock buffer) were measured with an oscilloscope. By switching off thePQR clock and switching it back on we could see the stabilization time of the PQR clockand how the XYZ’s LO phase and frequency adapted. After approximately 10-20 clockcycles (approximately 60 ns) the signals seemed to be in phase and frequency, howeverthis cannot count for a precise or very scientific method. We attempted to use vectorand frequency analyzers instead of the oscilloscope, however without any success.

We did not spend any time in optimizing the delay and offset for the SIG3 or SIG2signal. When optimizing these the XYZs configuration together with the SIG2 signalboth have to be taken into account as they share the same signal. An optimization ofthese might give even more savings since there might be more time to take advantageof it would also ensure proper functioning.

The strobed signal SIG1 during one TTI is shown in Figure 22. It has an delay, i.e.it turns on and off a bit later which must mean that the signal passes later here thanthrough the drivers. This makes sense because the signal controls the last amplificationin the amplifying chain. The maximum time it is possible for the PA and drivers tobe turned off is 10/14 = 71.43%, as at least four symbols are used to send CRS in idlemode. The sampling rate is 2.5MHz (2500 data points for one TTI) and to calculatethe time the PA is off all the measurement points above a 0.5V threshold are addedtogether. The result is 1784/2500 = 71.36% which more or less is exactly the maximumconcluding that the rise time for the PA is instantaneous.

Kent Persson told us in a telephone conversation that in the next generation of theASIC called ASIC2 there will be more strobing options and SPI commands will also bepossible to strobe. This means that on that platform TX micro sleep can be even moreeffective. It will be possible to include clock gateing and other components to the feature.

In Figure 23, showing the radio’s power consumption, the maximum output is the samefor Ericsson’s and our implementation, only the base power is lower. The reason isbecause all the components (SIG1, SIG2, SIG3 and XYZ) are up and running whentransmitting and turned off when not.

42

6.4 MIMO sleep

When transmitting from two antennas different measures of creating diversity is possiblewhich can improve the signal quality at the receiver by supplying two diversified signalssending the same information. The receiver then combines the two signals with ad-vanced signal processing to create one signal with better quality. Two antennas can alsobe used to create two parallel channels sending different information on two channelson the same frequency. This is called spatial multiplexing and allows for better use ofbandwidth, i.e. increases throughput. When the signal strength is good this is what thetwo antennas will be used. When the signal strength becomes very bad beam-formingshould be used instead to improve the signal. However this is not used, the modulationis degraded instead. So when turning off one branch the maximum throughput of theradio is lowered, but also better matched with the demanded throughput. Switching offon RX branch in addition will further reduce the power consumption. This will howeveraffect the uplink signal quality since the benefits of spatial diversity are lost. It is hardto predict the behavior of UEs and to determine that the signal strength is good enoughwith only one antenna might be difficult. To switch off only the components that comeback up fast, for example the LNAs, might be a possible solution since it allows for faston and off switching.

To be able to scale the available capacity with the traffic load makes sense becausethere is no need to have the whole radio up and running when only half supplies thedemanded capacity. In an area where there are periods of time of which the radio doesnot run at full capacity, this feature will reduce the power consumption dramatically.However there are some further improvements that can be done to the feature. Cur-rently it is not possible to gate the clocks for DL B in the ASIC without it failing, gatingthese clocks would give additional savings of around one watt.

The data for MIMO sleep is plotted in Figure 27. The pattern is identical, howeverthe overall power consumption has been reduced. This is due to that components notused in branch B are permanently turned off.

6.5 Combined features

The greatest reduction of power consumption is achieved when the TX micro sleep fea-ture is combined with the MIMO sleep feature. This reduces the power consumptiondramatically and takes it down to around 32W in Idle mode which is very close to whatthe radio consumes in Blocked Cell mode. In other words, we managed to reduce thepower consumption of the radio while still in operation to a level close to the currentEricsson implementation used to turn the radio off. We can now have the radio func-tioning normally however with reduced throughput. It consumes as little power as ifit was turned off in regards of what is possible today. The power consumption will ofcourse still increase with the load put on the radio and at full capacity the features willhave no effect and no savings will be possible.

This combined feature could be implemented on a large number of sites. In the be-ginning of the report Figure 4 shows how much traffic passes through sites and as wecan see half of them are low traffic sites. The combined feature would have a greatimpact on these sites since it scales with the demanded load. If there is low trafficthere are many symbols where the radio does not transmit and this means that thefeature has more active time. The capacity demand on the medium sites might be morethan what one antenna could provide however during some time periods the combinedfeature would be applicable. TX micro sleep will still be active and reduce the power

43

consumption. At the high traffic sites the power saving features will not have muchimpact as the radio is sending continuously. Only 10% of the sites are high traffic sitesso the remaining 90% could all reduce their power consumption significantly.

6.6 Scalability

During the study of the radio it became apparent that all components are always upand running, allowing for full capacity at any time, even though only a small percentageof the radio’s capacity is needed. The subject has been discussed earlier, to scale theavailable capacity of the radio to the demanded is clearly beneficial leading to a reducedpower consumption. The DU always knows exactly how much capacity is needed andwhen, making a scalable system possible. One way to do this is the MIMO sleep feature,we now have a system with two ”gears”, full and half capacity. We want to introducemore gears, we want a system that continuously changes gears depending on the demandof capacity.

One easy way to introduce gears would be to change the clock frequency for the ra-dio, when the traffic is low the clock frequency, or speed, of the system should be low.As the demand for capacity increases the radio switches gears; it increases the clockfrequency. If the frequency is lowered the voltage can be reduced, leading to additionalsavings. We researched the possibilities to reduce the frequency of the radio and weretold ”It is not possible to reduce the frequency.”, we were not deterred by this and kepttrying. However the task proved to cumbersome and was abandoned due to lack of time,however still convinced that it is possible. The radio itself can easily be clocked downbut when it runs on a different clock than the DU problems arise. This could howeverbe dealt with by increasing the clock speed in the components where those problemsoccur. These components have to be identified and the clock frequency determined.

The gears should also include components of the radio. The MIMO sleep feature for ex-ample include components that take about 500µs to turn on making it unfit for switchingon and off fast. By not dividing components into features and instead categorize themby their function and response time and asking the question ”What needs to be turnedon?”, a truly scalable system can be designed. This will be addressed more under 6.8.

6.7 Hardware

Many components come with the different power modes and some of these componentshave the pin to ground not allowing control and others can be activated by SPI or GPIO.To fully utilize the power saving capabilities of the components, they should not be hard-wired to ground and we need to be able to access them instantly. Hardware needs tobe modified to enable fast communication with the components whether it is by GPIOor SPI. Rapid interaction with the components enable them to be turned on and offfaster and therefore be utilized in shorter time windows. The ability to generate strobesfor all components on the board would be incredibly useful for many implementationssince it is known beforehand when the radio needs to transmit. Components should alsobe able to be controlled individually. In the current implementation many of the RXcomponents are controlled pairwise which makes scaling the RX chain impossible.

In general the components should be optimized out of a energy perspective with differ-ent power save modes. The XYZ has programmable pins which execute when the pin isset high, this allows for custom sleep modes to be constructed. This is useful when theinternal components have different response times and the preprogrammed sleep modes

44

are too slow to be utilized in short time frames. It is also desirable to have componentswhich have features that can be activated easily, by toggling a pin for example.

Looking at older radios, the RRUS 12 for example, many components cannot be con-trolled at all. The RRUS 12 has the same XYZ as the Micro but lacks the option tocontrol it by the pins PIN1 and PIN2 which we have used. The next generation radiois supposed to have more and faster controlability.

6.8 Making the radio intelligent

With all components categorized by their function and response time the question ”Whatneeds to be turned on?” can be answered by an intelligent system. The idea is to createa smart system, a type of artificial intelligence, that makes predictions based on his-torical data and current readings and then answers the question by turning everythingelse off. It will not scale by activation or deactivation of whole features but individualcomponents, and maybe not turn an component off entirely but put it in a sleep mode,depending on the current and predicted future demand. This allows for only the abso-lute necessary components to be active on any given time.

It seems preferable to put this intelligence in DU since it knows the current number ofUEs connected to which cell, it controls handovers and contains computational power.

45

7 Further work

Confidential.

8 References

References

[1] Miljo och energidepartementet. Mal for energi. 2015. url: http://www.regeringen.se/regeringens-politik/energi/mal-och-visioner-for-energi/.

[2] Tele2. Environmental Responsibility. 2016. url: http://om.tele2.se/miljo-och-hallbarhet/.

[3] Telia. Miljo och hallbarhet. 2015. url: http://www.teliacompany.com/en/

sustainability/responsible-business/environmental-responsibility/.

[4] Vodafone Germany. Energiesparmaßnahmen. 2016. url: http://www.vodafone.de/unternehmen/klimaschutz/energiesparmassnahmen.html.

[5] Deutsche Telekom. Mission: 20 percent fewer CO2 emissions. 2016. url: http://www.telekom.com/corporate-responsibility/climate-and-environment/

co2-reduction/64852.

[6] Vodafone Italy. Efficienza Energetica. 2016. url: http://www.vodafone.it/portal/Vodafone-Italia/Sostenibilita/Efficienza-energetica.

[7] Wind. Reducing atmospheric emissions. 2016. url: http://www.windgroup.it/eng/regolamentare/riduzione_emissioni.phtml.

[8] TIM. TIM SUPPORTS #GLOBALGOALS. 2016. url: http://in.tim.it/

iniziative/ambiente-e-sociale/.

[9] Confidental. -. -. url: -.

[10] Ericsson Press Backgrounder. Global Services – transforming industries with ICTexpertise. 2016. url: http://www.ericsson.com/res/thecompany/docs/press/backgrounders/global_services_press_backgrounder.pdf.

[11] Skold Dahlman Parkvall. 4G LTE/LTE-Advaced for Mobile Broadband. AcademicPress, 2011.

[12] Ericsson White Paper. LTE: an introduction. 2011. url: http://www.ericsson.com/res/docs/2011/lte_an_introduction.pdf.

[13] National Instruments White Paper. What is I/Q data? 2016. url: http://www.ni.com/tutorial/4805/en/.

[14] Chris Johnson. Long Term Evolution - In Bullets. Chris Johnson, Northampton,England, 2012.

[15] RTR. LTE Bands Overview. 2016. url: https://www.rtr.at/de/tk/FRQ_

spectrum/LTE_Bands_Overview.pdf.

[16] Confidential. -. -. -. -.


[18] Texas Instruments Edgar Pineda. Clocks Basics in 10 Minutes or Less. 2010. url:http://www.ti.com/ww/mx/multimedia/webcasts/TI_webinar_25- 06-

2010.pdf.


46

[20] Robert W. Erickson Dragan Maksimovic. Fundamentals of Power Electronics.Kluwer Academic Publishers, 2004.








47

9 Appendix

9.1 Abbreviations

3GPP 3rd Generation Partnership ProjectADC Analog to Digital ConverterASIC Application Specific Integrated Circuit

BE Back EndCPRI Common Protocol Radio InterfaceCRS Cell Specific Reference SignalingDAC Digital to Analog ConverterDPA Driver PADPD Digital Pre DistortionDSA Digital Step AttenuatorDU Digital Unit

FDD Frequency-Domain DuplexFE Front End

FPGA Field Programmable Gate ArrayGPIO General Purpose Input OutputLDO Low Drop Out regulatorLESS Low Energy Scheduler SolutionLNA Low Noise Amplifier

LO Local OscillatorLTE Long Time EvolutionLTU Local Timing Unit

LVDS Low Voltage Differential SignalLVPECL Low Voltage Positive Emitter-Coupled Logic

MIMO Multiple Output Multiple InputMPA Main PA

PA Power AmplifierPCB Printble Circuit BoardPLL Phase Locked Loop

QAM Quadrature Amplitude ModulationRAN Radio Access Network

RX ReceiverSERDES Serializer/Deserializer

SFP Small Form-factor PluggableSPI Serial Peripheral Interface

TDD Time-Domain DuplexTRX TransceiverTTI Transmission Time IntervalTX TransmitterUE User Equipment

VCO Voltage Controlled Oscillator

Table 16: Commonly used abbreviations.

48

Energy Optimization of Radio - Lu

Documents