On performance of LTE UE DFT and FFT implementations in flexible software based baseband processors

The use of the Fast Fourier Transform (FFT) and the Discrete Fourier Transform (DFT) in the physical layer implementations of the latest cellular and wireless standards such as Long Term Evolution (LTE) is highly common place. Given that FFT and DFT algorithms consume a significant percentage of the processing power of a baseband processor, it is important to implement these algorithms with close attention to the underlying architecture of the platform for which they are being developed. In this paper novel approaches to implementation of various sizes of the DFT required in the LTE standards are analyzed and benchmarked. In particular, implementation of the algorithms is assumed to be based on a flexible software based baseband processor. This platform therefore requires the DFT implementation to be completely in software. This poses an important constraint, as typical hardware based implementations use very specific hardware architectural features to minimize the computational latencies. On the other hand a software implementation requires utilization of the general purpose (i.e. load/store) instructions of a processor, as well as its specialized instructions (e.g. butterfly and complex multiplications). In addition transfer of massive amounts of data to and from the processor need to be addressed. This requires proper usage of DMA (Direct Memory Access) where appropriate. Data storage requirements are also an important consideration in software implementation of DFT algorithms. Software based platforms have typically limited on chip (fast) memory, that may restrict storage of butterfly weights and other parameters required to set up a DFT process. The Third Generation Partnership Project (3GPP) has been defining the Long Term Evolution (LTE) for 3G radio access. LTE has several areas of focus. These areas include enhancement of the Universal Terrestrial Radio Access (UTRA), as well as optimization of the network architecture with HSDPA (downlink) and HSUPA (uplin- k). LTE project aims to ensure the continued competitiveness of the 3GPP technologies for the future LTE focuses on download rates of 100 Mbit/s, upload rates of 50 Mbit/s per 20 MHz of bandwidth, increased spectrum efficiency and sub-5ms latency for small IP packets.