Adaptive cellular telephony coprocessor

The computational requirements imposed by cellular telephony standards necessitate more performance than is currently available on embedded processors. The problem will persist in the future since algorithmic complexity is growing faster than Moore's Law. The additional challenge to embedded processors is how to meet these performance needs under stringent energy budgets. The rapid rate of change in cellular telephony requirements motivates the creation of new architectures which are also flexible. Achieving high-performance and flexibility within a limited power budget is a difficult task.; The traditional approach for applications requiring both high-performance and low power is to employ dedicated integrated circuit solutions for compute intensive components. In areas where applications evolve rapidly, flexibility is also an important factor, and a general purpose or embedded processor approach has often been used for this reason. Moore's Law scaling has allowed embedded processors to become more powerful. This improvement is insufficient for portable computing and communication devices that increasingly employ rapidly evolving, sophisticated and power hungry signal processing techniques. Moreover, these processors incur a significant overhead in terms of power and area compared to specialized hardware. This is caused by excess die area being consumed for control functions which leaves insufficient resources for data path execution duties.; These problems motivate the investigation of an alternative approach. In this dissertation, the approach is to trade programming complexity for efficiency in order to reduce hardware complexity and explicitly control the communication resources. The idea is to design a processor with a reconfigurable data path where special purpose computational pipelines can be dynamically established that resemble data flows found in the dedicated circuit solutions. These pipelines can be instantly reconfigured to support a new processing phase. By executing these algorithms on dedicated pipelines, significant energy savings and performance improvement is possible. Flexibility is achieved by fine-grained program control of the communication between the pipeline stages.; The main contribution of this work is a low power flexible architecture for adaptive cellular telephony (ACT) algorithms. The architecture is flexible enough to efficiently support diverse algorithms. The effectiveness of the design, in terms of power, flexibility, and performance, is evaluated on algorithms taken from the wireless domain and generic DSP kernels. ACT is within one order of magnitude of the energy-delay product of an application specific integrated circuit and three to four orders of magnitude more efficient than a low power embedded processor implementation.