Adapting the LHC 1TFB electronic circuit to other equipments The candidates are: PS 1TFB PS TFB PS CBFB PSB TFB PSB 1TFB 1 Alfred Blas Working group meeting - 26 March 2009 1. Clock frequency range and related issues 2. Required Delayed Clocks 3. Secondary effects of the fine delays switching 4. Stability analysis of the PS 1TFB without a Notch Filter
15
Embed
Adapting the LHC 1TFB electronic circuit to other equipments The candidates are: PS 1TFB PS TFB PS CBFB PSB TFB PSB 1TFB 1 Alfred Blas Working group meeting.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Adapting the LHC 1TFB electronic circuit to other equipmentsThe candidates are:PS 1TFBPS TFBPS CBFBPSB TFBPSB 1TFB
1Alfred Blas
Working group meeting - 26 March 2009
1. Clock frequency range and related issues
2. Required Delayed Clocks
3. Secondary effects of the fine delays switching
4. Stability analysis of the PS 1TFB without a Notch Filter
1.Clock frequency range and related issues
The most demanding equipment in that respect is the PSB TFB
2Alfred Blas
Working group meeting - 26 March 2009
PSB revolution frequency with Linac2 [ 0.6 MHz , 1.73 MHz ] (factor 2.88)
PSB revolution frequency with Linac4 [ 1 MHz , 1.73 MHz ] (factor 1.73)
Sampling rate for having a 20 MHz system analogue bandwidth > 60 Ms/s
Present ADC (AD6645) clock frequency range: [ 30 MHz , 105 MHz] DAC (AD9754) clock frequency range: [ DC , 125 MHz]
Final limitation for the sampling frequency : 60 MHz < fS < 105 MHz
The ratio 105/60 = 1.75 > 1.73 is compatible with a use in the PSB during the Linac4 era !
< 2.88 means it is not compatible with Linac 2!
For a use within the Linac 2 era either:1. Change the ADCs / DACs2. Change the clock harmonic during the accelerating cycle
1.Clock frequency range and related issues
The most demanding equipment in that respect is the PSB TFB
3Alfred Blas
Working group meeting - 26 March 2009
ADCs can be found with 16 bits and a clock frequency range = [ 30 MHz , 160 MHz ]
Nevertheless there are at reasons not to change them:
1. It requires resources that are lacking2. It would be only useful in the frame of Linac 2 which will be
replaced in 20133. Keeping a high sampling rate by the mean of a clock harmonic
change is much more efficient as it allows to keep a higher analogue bandwidth with the same hardware
To be able to change the clock harmonic on the fly, the following conditions need to be fulfilled:
4. Clock source and sampled device need to be synchronous for the change of harmonic
5. The new harmonic value needs to be known (to allow the proper change of filters and delay parameters)
6. The loop signal will not be available for the duration of the purging of the old data sampled with the previous clock harmonic ( 1 revolution). This shortcoming is not inevitable but is of little effect and allows a simplification of the hardware.
1.Clock frequency range and related issues
The most demanding equipment in that respect is the PSB TFB
4Alfred Blas
Working group meeting - 26 March 2009
For the harmonic change to be synchronous on both the sending and receiving ends:
1. The clock can be tagged… but the hardware would have to be modified at both ends
2. A special trigger could be send in parallel with the clock signal (a: with a double twisted pair or b: with two coax cables of the same length). Prop. a means hardware and cabling changes; prop. b means to install new cables and doesn’t give precise results (although we could live with it (see below).
3. The clock frequency could be measured online and when an abrupt change is detected the new harmonic is applied.
For the new harmonic value to be known:
4. The hclock values could be sent from the clock source as a serial word. This solution requires hardware changes and is heavy.
5. The hclock values could be sent as the content of registers. The first register giving the value at injection and the others the following values. The central control system would update the registers on both ends on a ppm basis
6. The hclock value could be measured online, but this is time consuming.
7. The hclock values could be pre-determined once forever. Always the same value at injection; always the same clock frequency ratio at each change. This solution means no communication.
1.Clock frequency range and related issues
The most demanding equipment in that respect is the PSB TFB
5Alfred Blas
Working group meeting - 26 March 2009
The proposed solution for the clock range issue is:
1. Change the clock harmonic on the fly
2. Have the sequence of harmonic values programmed in registers
3. Detect the harmonic change in real-time from a measurement at the receiving end. The detector response time will mean that corrupted (wrongly sampled) data will be sent to the hardware; this is acceptable if this response time is smaller than a couple of clock periods
4. Dump the invalid loop values acquired at the previous clock harmonic at the cost of no data for 1 turn
2. Required delayed clocksThe most demanding equipment
in that respect is the PS TFB
6Alfred Blas
Working group meeting - 26 March 2009
DPRAM
Ck
Ck + calculated
Δt
Data in
WriteAddres
s
Data out
ReadAddress
Counter
Calculated pipeline delay
Ck
2 different clock domains !
Write
Read
One delay needs to be applied to equalize the delay between 2 PUs; another needs to be applied to adapt the total loop delay.
2. Required delayed clocksThe most demanding equipment
in that respect is the PS TFB
7Alfred Blas
Working group meeting - 26 March 2009
When ΔT < tpd + th the read address needs to be latched with CK (tpd = CK to data out propagation delay, th = flip-flop hold time)
When tpd + th < ΔT < TCK the read address needs to be latched with /CK (inverted CK)
The diagrams on this page represent the situation with ΔT = 0 ns when the fine delay is programmed to zero (clock and delayed clock are in phase). This is not the case presently.
2. Required delayed clocksThe most demanding equipment
in that respect is the PS TFB
8Alfred Blas
Working group meeting - 26 March 2009
If ΔT was not equal to 0 ns when programmed to zero (actual case = 8.8 ns) the strategy would be more complicated and would require the knowledge of the offset delay. Value to be available in a data-base. => complicated handling.
2. Required delayed clocksThe most demanding equipment
in that respect is the PS TFB
9Alfred Blas
Working group meeting - 26 March 2009
To ease the operation of the board, he following setup is recommended:
ΔToffset ΔToffset
ΔTvar ΔToffset
ΔTvar
CK
CK + ΔT1
CK + ΔT1 + ΔT2
Rev clock
10Alfred Blas
Working group meeting - 26 March 2009
2. Required delayed clocksADC Data synchronization
ADC conversion time: 1.4 tC 7ns
FPGA flip-flop hold time = 1ns?
Longer path in the FPGA from pin to F-F compared pin to clock path :3ns?
2. Required delayed clocks
11Alfred Blas
Working group meeting - 26 March 2009
ΔToffse
t
ΔToffse
t
ΔT1
ΔToffse
t
ΔT2
CK
CK + ΔT1
CK + ΔT1 + ΔT2
Rev clock
ΔTADC
ΔTADC
ΔTADC
ADC
ADC
Data ADC 1
Data ADC 2
/CK
/(CK + ΔT1)
/(CK + ΔT1 + ΔT2)
3. Effect of a delay change
12Alfred Blas
Working group meeting - 26 March 2009
DPRAM
Ck
Ck + calculated
Δt
Data in
WriteAddres
s
Data out
ReadAddress
Counter
Calculated pipeline delay
Ck
Write
Read
During an accelerating cycle the automatic delay will decrease in the following way:
Smooth decrease of the fine delay Δt down to zero -> then decrease of one pipeline stage together with an abrupt increase of the fine delay Δt of about one clock period.
Opposite behavior in a decelerating cycle.
3. Effect of delay change
13Alfred Blas
Working group meeting - 26 March 2009
This delay transition (decrease of one pipeline stage) should be smooth with no glitch. The memorized clock signal in the fine delay total length has no side effect (assuming a smooth functioning of the file delay chip).
4. Analysis of 1TFB stability without the Notch Filter
14Alfred Blas
Working group meeting - 26 March 2009
The analysis will be made with a gain compensation in the forward AVC loop path to compensate for the reduction due to the presence of a gain at the rf frequency in the feedback path.
To be completed !This analysis will have no consequences on the PCB architecture.
4. Summary Things to be taken into account
for the hardware development
15Alfred Blas
Working group meeting - 26 March 2009
1. The ADCs should be kept as they are (although the 105 MHz version is preferred)
2. The clock distribution should be implemented as described in slide 11.
3. The smooth functioning of the fine delay chip should be checked for (no missing or extra clock rising edge when changing the delay value)
4. The clock reception unit should be designed to deal with distant signals (ground decoupling)