130 likes | 280 Views
Architectures for Baseband Processing in Future Wireless Base-Station Receivers. Sridhar Rajagopal ECE Department Rice University March 22,2000. This work is supported by Nokia, Texas Instruments, Texas Advanced Technology Program and NSF. Third Generation Wireless. First Generation
E N D
Architectures for Baseband Processing in Future Wireless Base-Station Receivers Sridhar Rajagopal ECE Department Rice University March 22,2000 This work is supported by Nokia, Texas Instruments, Texas Advanced Technology Program and NSF
Third Generation Wireless First Generation Voice Eg: AMPS Second/Current Generation Voice + Low-rate Data (9.6Kbps) Eg : IS-95(N-CDMA) Third Generation + Voice + High-rate Data (2 Mbps) + Multimedia W-CDMA CAIN Project
Noise +MAI Base Station Reflected Paths Direct Path User 1 User 2 Main Parts of Base-Station Receiver • Channel Estimation • Noise, MAI • Attenuation • Fading • Detection • Detect user’s information • Multiple Users • Decoding • Coding/Decoding improve error rate Performance • Coding done at handset Wireless Communication Uplink CAIN Project
User Interface Translation Synchronization Transport Network Physical Layer OSI Antenna Layers 3-7 Multiuser Detector Data Decoder Demux Data Link Layer (Converts Frames to Bits) OSI Layer Estimated Amplitudes & Delays 2 Pilot Channel Estimator Physical Layer (hardware; raw bit stream) OSI Layer 1 Base-Station Receiver CAIN Project
5 x 10 Data Rates for a typical DSP Implementation 2 Data Rates 1.5 Data Rate Requirement = 128 Kbps 1 0.5 0 9 10 11 12 13 14 15 Number of Users Need for Better Architectures • Current DSPs need orders of magnitude improvement to meet real-time requirements. • Reason • Sophisticated Algorithms, Computationally Intensive Operations • Floating Point Accuracy • Solution • Try sub-optimal/iterative schemes • Fixed Point Implementation • Use structure in the algorithms • Parallelism / Pipelining • Task Partitioning • Bit Level Arithmetic CAIN Project
Channel Estimation - An example • Channel Estimation† includes • Matrix Correlations, Matrix Inversions, Multiplications • Floating Point Accuracy • Need to wait till all bits are received. • Modified Channel Estimation Algorithm • Matrix Inversion eliminated by Iterative Scheme • Based on Gradient / Method of Steepest Descent • Negligible effect on Bit error Performance • Fixed Point accuracy, Computation spread over incoming bits • Features to support Tracking over Fading Channels easily added. †Maximum Likelihood Based Channel Estimation [C.Sengupta et al. : PIMRC’1998, WCNC’1999] CAIN Project
Comparison of Bit Error Rates (BER) -1 10 -2 BER 10 O(K2N) MF ActMF ML ActML O(K2NL) -3 10 4 5 6 7 8 9 10 11 12 Signal to Noise Ratio (SNR) Simulations - AWGN Channel Detection Window = 12 SINR = 0 Paths =3 Preamble L =150 Spreading N = 31 Users K = 15 10000 bits/user MF – Matched Filter ML- Maximum Likelihood ACT – using inversion CAIN Project
DSP Implementation • Advantages • Programmability • Ease of implementation • High Performance • Low Cost • Disadvantages • Improvements necessary to meet real-time requirements! • Sequential Processing • Parallelism not fully exploited • Cannot process or store data at granularity of bits. CAIN Project
VLSI Implementation • Task Partition Algorithm into Parallel Tasks • Take Advantage of Bit Level Operations • Find Area-Time Efficient Architecture • Meets Real-Time Requirements! Task A Task C Task B Time CAIN Project
Conclusions • Better Performance achieved by • Modifications in the Algorithm • Application Specific Architectures • Algorithmic Modifications • reduce the complexity of the algorithms • develop sub-optimal or iterative schemes. • Custom hardware solutions • bit level operations and parallel structure. • Together, algorithm simplifications and custom VLSI implementation can be used to meet the performance requirements of the Base-Station Receiver. CAIN Project
Future Work • Analysis for Detection and Decoding • Mobile Handsets • Mobile handsets have similar algorithms • Need to account for POWER too. • General Purpose Enhancements [But, VLSI first ] • Explore Instruction Set Extensions / Architectures for DSPs • Exploit Matrix Oriented Structures • Bit Level Support • Complex Arithmetic CAIN Project
0 10 MF - Static MF - Tracking ML - Static ML - Tracking -1 10 BER -2 10 -3 10 4 5 6 7 8 9 10 11 12 SNR Fading Channel with Tracking Doppler Frequency = 10 Hz, 1000 Bits,15 users, 3 Paths CAIN Project
Talk Outline • Introduction • Need for better Architectures • Channel Estimation - An example • Simulation Results • Implementation Issues • General Purpose/Application Specific • Conclusions • Future Work CAIN Project