GraduateCategory: Engineering and TechnologyDegree Level: Ph.D. Abstract ID# 1112

Energy-Efficient Wireless Transceiver Operation using Optimized Heterogeneous Processor-FPGA Computations

Benjamin Drozdenko, Matthew Zimmermann, Tuan Dao, Kaushik Chowdhury, Miriam LeeserAbstract

• Recent explosion in the number & diversity of devices, protocols, and applications• Trend points towards systems with high data rates and low energy consumption• Emerging vision of creating a transceiver architecture that can adapt to the

functional and processing needs of existing and future protocols• Maps computation to underlying heterogeneous platform, with CPU and FPGA• We introduce a method for modeling 802.11a-based OFDM wireless transceiver• Prototype on Xilinx Zynq system-on-chip by dividing PHY layer into functional units• Our approach creates MathWorks Simulink model variants for both transmitter

and receiver, each with a different boundary between HW and SW components• Use models to generate HDL code-FPGA bitstream & C code-ARM CPU executable• Results demonstrate how to select a HW-SW codesign for ideal 802.11a operation

Background

• Design follows IEEE 802.11a Physical (PHY) layer specifications [1]• Scrambling: XOR to make data pseudorandom, unbiased, independent• Interleaving: rearranges bit indices to make random errors seem more random• Convolutional Encoding: adds redundancy by producing parity bits• Binary/Quadrature Phase Shift Keying (B/QPSK): modulation from bits to symbols• Orthogonal Frequency Division Multiplexing (OFDM): map symbols to subcarriers

and use Inverse Fast Fourier Transform (IFFT) to carry data on multiple channels• Preamble: the fixed initial

sequence at the start of atransmission, used to detect the beginning of frame at Rx

Conclusion

• For direct feedthrough algorithms, moving more components to execution in HW results in faster execution speed, but adds risk of overwhelming FPGA resources

• While energy consumption increases as more components are placed on PL, the amount is negligible when compared to the embedded ARM energy consumption

• Many of the components developed for this base design can be reused for other variants of 802.11 (ac, af) as well as LTE protocols (4G mobile)

• For future work, we plan to perform tests with online radio transmissions and measure bit error rate (BER) for the different HW-SW co-designs

References

[1] IEEE 802.11 Working Group, “IEEE Std802.11a-1999.” IEEE, 1999.[2] Analog Devices, Inc. (2015) Integrated transceivers, transmitters, and receivers. [Online]. Available: http://www.analog.com/en/products/rf-microwave/[3] MathWorks, Inc. (2015) Xilinx Zynq Support for MATLAB and Simulink. [Online]. Available: http://www.mathworks.com/hardware-support/zynq.html [4] G. Eichinger, K. Chowdhury, and M. Leeser, “CRUSH: cognitive radio universal software hardware,” in 22nd FPL, Oslo, Norway, August 29-31, 2012[5] J. Pendlum, M. Leeser, K. Chowdhury, “Reducing processing latency with a heterogeneous FPGA-processor framework,” IEEE FCCM 2014, Boston, May 11-13, 2014[6] B. Drozdenko, R. Subramanian, K. Chowdhury, M. Leeser, “Implementing a MATLAB-based self-configurable software defined radio transceiver,” 10th CROWNCOM ‘15, Doha, Qatar, Apr 21-23, 2015

Results: Execution Time

• Faster speed moving more components on PL. IFFT & Preamble Detection longest.

Hardware Components

• Platform uses Xilinx ZC706 Evaluation Kit, ADI FMComms3 RF front end, Host PCSoftware Tools

• Workflow uses MathWorks Simulink, HDL Coder, Embedded Coder, Xilinx Vivado

Method: Hardware-Software Codesign Model Variants

• Create seven models to represent HW-SW divides between 802.11 function units Hardware-Software Interfacing

• AXI-stream interface uses DMA controller to transfer data between PS & PL

Results: Resource Utilization and Energy Consumption

• As more functional units put on PL, utilization & energy usage gradually increases

Acknowledgements

Discussion

• A benefit of flexible SDR testbed is reuse for other 802.11 & mobile standards• Some units (scrambling, interleaving) can be reused directly in their present form• Modifications needed for different encoding rates (2/3) and modulation schemes• This reusability allows us to explore LTE and Wi-Fi coexistence on the same

channel, TV whitespace reuse, and co-operation with RADAR

PL Data Path Delay PL Energy Usage