Baseband Symbol Processor For Micro Base Stations In 5G Ultra-dense Network

In this research,we studied on baseband system design for low-cost UDN micro base stations.Baseband chip-set solutions based on software-defined radio(SDR)are proposed,and an application specific instruction-set processor(ASIP)for UDN symbol processing is designed.Key innovative contributions of this research are as follows.First,top-level power and cost estimation models are proposed based on the static hardware modeling method.At the 5G pre-study stage when dynamic method cannot be adopted,this modeling method can be used for early power and cost estimation of the UDN baseband,and can guide the design and exploration of UDN baseband chips.Second,comprehensive and quantitative analyses are performed for the digital baseband of 5G UDN base stations,including the function flow definition,the selection of baseband algorithms,the analysis on algorithm complexities,and the analysis on implementation challenges.Especially,the featuring algorithms of 5G: massive-MIMO beamforming algorithms,is analyzed.Based on trade-off between performance and complexity,algorithm selection recommendations are provided depending on antenna number and communication scenarios.For other algorithms,a scheme is given with fully-defined functions,data width,complexities and latencies.Third,design and planning of low-cost baseband chip-set is realized.We provide method and solution to the chip partition problem targeting low cost under pin and power constraints.Based on this method,we fully explored chip-set solutions under different BS configurations,different I/O schemes,and different beamforming schemes.Power and cost analysis results are given.Recommendations are then given on architecture implementations of UDN baseband.Finally,design of the ASIP architecture for UDN baseband symbol processing algorithms is carried out.On the SIMD processor,instruction set is designed for efficient acceleration of symbol processing algorithms,and flexible addressing is designed for low overhead vector-based data access.Algorithm mapping of symbol processing kernels are done based on assembly coding,and performance evaluations are provided.Results show that for kernels with medium/high data parallelism,the overhead of processor execution has approached the theoretical limit that we have deduced for the evaluation.Academically,the main visions and conclusions in this thesis are: first,based on the exploration of system parameters,it is proved that antenna number is the factor impacting power and cost in the first place,while communication scenarios and mobility are not key factors;second,based on cost and power estimation results,it is concluded that the benefit from hybrid beamforming structures is undesirable.On engineering,the advanced contributions are: first,the architectural uniformity of baseband in 5G base stations is discussed,and the feasibility of SDR based baseband system for 5G base stations is proved;second,the complexity analysis,architecture proposal,cost and power estimations for the baseband of UDN base staion at hardware level is accomplished for the first time.