Proceedings of ITC-CSCC ¯97 Okinawa, Japan Design of A Vita -bi Decoder with Sequence- Processing CapabHity Jot- -M oon Choi, Jung-n Han, W oo-Young Choi, Borg ¤ yul U rn Dept. of Electronic Engineering, Yonsei University 134, ShHU M ID or- , Sudae m on- Ku, Seoul 120 %9, Korea E- md : humph@semicord a- rIsa ac-kr Fa : +a z a l- -a Abstract : In this paN r, we proposed a Vherbi decoder for practical implementation of For ward E- or Correction- EC). The design was ma nly f- used on real- time prÄ essing by using the memory organization m d its control method with cons- a nt length 7, coding rate 1/ 2. M anagement of memory contents in a Vita -u decoder is a major design problem for both hardware and software realization. In this design, we solved that problem with sequence processing. 1. Introduction h ma y d gital comfy-lullcanon systems, etTor com- don circuitry is widely used in order to reduce the bit erTOr- rate of u ansmit- d data [2,5]. Convolutional coding with Vita -bi decoding is a powerful method for forward error correction [ 1- 5] . In present commercial applications, V ita -u decoder s with R(coding rate)=1/2 a d K(constraint length)=7 are widely used. A very simple v an ation of the Vitemi algorithm permits t he u se of soft-da ls on demodulated data in which signal values a - quantized into multiple levels a d digitized [2,5]. The advantage of soft decision demodulation 1S mat the signal values not only indicate whether they represent one or zero, but also indicate £ e magnitude of the con-t»ption of the signal by noise at the instant of quanna tion- A significant processing ga n can be achieved by using soft da ismn data, typical- about 2 dB for 3- bit dat a. We designed a Vherbi decoder mat has R=1/2, K=7 m d 3- bit soft decision. Our design has a trace- back a clutc h- e ad achieves Va ir- , uM ating a d stou r- of the real- time sequences within a single clx k cycle. 2 . V h er bi A 1g o¢ £ m T he Vita -M algorh hm @Horn½S the maximum likelihood decoding [ 1- ð T he adv¤ ttage of Vitemi decoding is mat the complexity of the decoder is not a function of the number of symbols in the codew or d sequence [2]. T he algor1thm determines a measure of similarhy between received signals, at time ti, a d d l the trellis paths entering each state at time ti. Then, it removes V eHi s paths m at cm not be a ca didate for the maximum likenhÄ d choice. When two paths enter the sa ne state, me one having the smallest meM c is chosen. T his path is caned the survivor path. The M ecum of survivor path is * rformed for all the states. T he da Oder- continues in this way a d makes decisions by eliminating the least likely paths [2- 5]. T he early M a nort of the unlikely paths reduces the da Oding complexity [2]. Vita- i da Oder has three major pa ts. The first pa t calculates bra ch meM cs. The branch men-c is the likelihood meM c of ra eived cd es, and is calculated based on the comparison betw een the M lHs a d received convolutional codes [2- 5] . T he second pa t is ±A dd- Compare- Select (ACS)± pa t . In A CS, path ma rtcs are updated by adding bra ch me- -Cs associated with each possible state Vm o tion [6]. T he number of states N S of a convolutional encoder which generates n encoded bits is a function of the cons- a nt length K a d input bits b [1,2]. N . = 2b(KO (1) R = b/ n (2) T he £ ó d pa t per forms pat h seIa Oon and memery manager- nt. The smaller path metric is the path meM c for the state and me resultmg - 701