Optimization of advanced telecommunication algorithms from power and performance perspective

Abstract

This thesis investigates optimization of advanced telecommunication algorithms from power and performance perspectives. The algorithms chosen are MIMO and LDPC. MIMO is implemented in custom ASIC for power optimization and LDPC is implemented on dynamically reconfigurable fabric for both power and performance optimization. Both MIMO and LDPC are considered computational bottlenecks of current and future wireless standards such as IEEE 802.11n for Wi-Fi and IEEE 802.16 for WiMax applications. Optimization of these algorithms is carried out separately. The thesis is organized implicitly in two parts. The first part presents selection and analysis of the VBLAST receiver used in MIMO wireless system from custom ASIC perspective and identifies those processing elements that consume larger area as well as power due to complex signal processing. The thesis models a scalable VBLAST architecture based on MMSE nulling criteria assuming block rayleigh flat fading channel. After identifying the major area and power consuming blocks, it proposes low power and area efficient VLSI architectures for the three building blocks of VBLAST namely Pseudo Inverse, Sorting and NULLing & Cancellation modules assuming a 4x4 MIMO system. The thesis applies dynamic power management, algebraic transformation (strength reduction), resource sharing, clock gating, algorithmic modification, operation substitution, redundant arithmetic and bus encoding as the low power techniques applied at different levels of design abstraction ranging from system to architecture, to reduce power consumption. It also presents novel architectures not only for the constituent blocks but also for the whole receiver. It builds the low power VBLAST receiver for single carrier and provides its area, power and performance figures. It then investigates into the practicality and feasibility of VBLAST into an OFDM environment. It provides estimated data with respect to silicon real estate and throughput from which conclusion can easily be drawn about the feasibility of VBLAST in a multi carrier environment. The second part of the thesis presents novel architectures for the real time adaptive LDPC encoder and decoder as specified in IEEE 802.16E standard for WiMax application. It also presents optimizations of encoder as well as decoder on RICA (Reconfigurable Instruction Cell Architecture). It has searched an optimized way of storing the H matrices that reduces the memory by 20 times. It uses Loop unrolling to distribute the instructions spatially depending upon the available resources to execute them concurrently to as much as possible. The parallel memory banks and distributed registers inside RICA allow good reduction in memory access time. This together with hardware pipelining provides substantial potential for optimizing algorithms from power and performance perspectives. The thesis also suggests ways of improvements inside RICA architecture.