TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

FPGA Implementation Issues of a Flexible Synchronizer Suitable for NC-OFDM-Based Cognitive Radios

Tutkimustuotosvertaisarvioitu

Yksityiskohdat

AlkuperäiskieliEnglanti
Sivut102-116
JulkaisuJournal of Systems Architecture
Vuosikerta76
Varhainen verkossa julkaisun päivämäärä16 marraskuuta 2016
DOI - pysyväislinkit
TilaJulkaistu - toukokuuta 2017
OKM-julkaisutyyppiA1 Alkuperäisartikkeli

Tiivistelmä

This paper presents a flexible timing synchronization scheme alongside the hardware implementation issues on an Altera Stratix-V Field Programmable Gate Array (FPGA) device. The core content of the synchronizer is based on Finite Impulse Response (FIR) filter which operates as a multicorrelator on demand. The term “flexibility” refers to a specific part of the synchronizer where the multicorrelator reconfigures its FIR filter block on-the-fly by employing Partial Reconfiguration (PR) feature. Moreover, different implementations have been evaluated for the multicorrelator, including MultiplierLess (ML) approach (an approximate computing technique only for autocorrelation purpose) along with the Direct From as well as the Transposed, Parallel, and Pipelined-Parallel Direct Form FIR filters. All the developed architectures are compared to each other in terms of power consumption, silicon area, and maximum frequency. Preliminary synthesis results show that the ML approach achieves better performance (including 94% less power dissipation, 75% less logic utilization as well as 67% fewer registers) than other architectures when performing autocorrelation function. Furthermore, the critical path is analyzed and appropriate optimization techniques (such as DSP register packing and intermediate register insertion) are applied to the best candidates of the architectures mentioned. As the best results, 2.83× speed-up, 56.57% less logic utilization along with 38.86% fewer registers are achieved for different architectures. Accordingly, we discover that the parallel form, as well as the pipelined-parallel one, achieve more interesting results than the transposed version in most of the cases.