Baseband Signal Compression in Wireless Base Stations Aida Vosoughi, Michael Wu, and Joseph R. Cavallaro {vosoughi, mbw2, cavallar}@rice.edu Unused Significant USBR code Motivation 010, 5(=101) Experimental Setup and Results Bit Removal binary code 01011001 11001 Using WARP hardware and WARPLab framework for over-the- With new wireless standards, base stations require 01010000 10000 Used before in air experiments: (a) Downlink, (b) Uplink greater data rates over the serial data link between base 01000111 00111 Network on chip serial station processor and (remote) RF unit. 01001100 01100 link compression 10011100 1001, 4(=100) This link is especially important in: 10010011 1100 10011010 0011 • Distributed Antenna Systems 10010001 1010 • Cooperating base stations …. 0001 • Cloud-based Radio Access Networks … We propose to use a differencing step before compressing with Baseband signals: each of the above lossless schemes. • Usually oversampled and correlated Compression Ratio Results: • Represented with extra bit resolution Proposed Lossy Compression Lossless Schemes: Lossy Scheme: LSB Removal/Quantization: Algorithm TX RX DQPSK 16-QAM Compression can decrease hardware complexity and cost. The number of LSBs that can [1] 1.51 1.07 TX 4 2.3 be removed without causing USBR 1.41 1.17 RX 3.5 2 distortion depends on the Elias-gamma + 1.40 0.84 Introduction and Related Work Adaptive AC modulation type. Adaptive AC 1.26 0.95 Base station architecture with compression/decompression: Elias-gamma 1.22 0.72 With higher size constellations, more bit Bit resolution: DAC:16 bits, ADC: 14 bits resolution is needed less can be ignored. Downlink Uplink QPSK: Example of quantization compression/decompression: Recently proposed approach [1]: Transmitting 9 Compressor bit code-words Decompressor Binary code (16 bits) (# LSBs ignored = 7) Quantized code 0101-1001-0101-0000 0101-1001-0 0101-1001-0000-0000 0100-1100-1001-0110 Differencing decreases the range of the values: 0100-1100-1 0100-1100-1000-0000 0011-1001-0110-0111 0011-1001-0 0011-1001-0000-0000 1100-1000-1001-1001 1100-1000-1 1100-1000-1000-0000 16-QAM: …. …. … Bit Packing: The proposed lossy scheme can be implemented efficiently in hardware as it only needs bit shift operations. Simulation Results [1] A. Wegener, “Compression of baseband signals in base transceiver systems,” US Patent 8,005,152, August 2011. Studied Lossless Schemes Conclusions and Future Work Arithmetic Coding: Elias Gamma Coding: While lossless methods have the advantage of not • Entropy code • A prefix-free universal code introducing error at all, we show that quantization can • Alphabet {A,B} with distribution • Appends zeros before the have better compression performance without adding (p , p ) binary representation: A B noticeable distortion. • AC maps S= s1s2…sN, si ∈ {A, B} Implied to a half-open interval [x,y). Number Encoding probability • Example: Lossless schemes are not effective in uplink. 1 = 20 + 0 1 1/2 S = ABB, pA = 2/3, pB = 1/3 1 2 = 2 + 0 010 1/8 Future Work: 3 = 21 + 1 011 1/8 4 = 22 + 0 00100 1/32 5 = 22 + 1 00101 1/32 Adaptive compression: Different compression algorithms 6 = 22 + 2 00110 1/32 based on data rate requirements, SNR, … 7 = 22 + 3 00111 1/32 8 = 23 + 0 0001000 1/128 Asymmetric compression: RF module is generally desired 9 = 23 + 1 0001001 1/128 to be simpler than base station processor; uplink and 10 = 23 + 2 0001010 1/128 downlink differences. … … … .