Software Defined Radio (SDR) on Graphic Processing Unit (GPU)

Software defined radio (SDR) is quite useful as a flexible implementation of WLAN transceiver in wireless networks. It provides a programmable testbed for innovations in both Medium Access Control (MAC) and physical (PHY) layer network protocols, whereas traditional simulation based method is considered broken in terms of its reality. SDR gives researchers full control over the network protocol stack, as a common drawback for software based approach, the speed of SDR could not satisfy nowadays high-bandwidth wireless networks, e.g. 802.11 networks and Ultra-Wideband (UWB).

Software defined radio can also provide users flexibility in choosing communication protocols. For example, a cell phone equipped with SDR can be programmed to either GSM or CDMA depending on the networks. Another example is in wireless networks, devices equipped with SDR can be programmed to either Bluetooth, WiFi, UWB, etc,
depending on the users' requirements such as transmission range and data rate. For these cellular and wireless devices, using fast CPUs for signal processing is to costly and consumes too much power.

In this project, we plan to take advantage of the data parallelism of GPU, explore and evaluate the idea of porting SDR to GPU, both of which have a similar function of signal processing that can be efficiently parallelized. And we hope that the GPUs today with general-purpose oriented design could handle the task of SDR reasonably well.

Plan Of Attack:

1. Profiling GNU Radio OFDM on CPU
2. Port OFDM encoding/decoding code to GPU
3. Performance evaluation

Project Proposal

Project Milestone