Baseband Transceiver Design for the IEEE 802.16a OFDM mode Advisor : Tzi-Dar Chiueh Student : Sang-Jung Yang Date : December 15th , 2003

Outline Review of 802.16a System Channel Model Transceiver Architecture Coarse Symbol Boundary Detection Fractional and Integer part CFO Estimation Tracking Residual CFO Tracking TFO Encountered Problem Conclusion Reference

Scope of 802.16a (1/3) 802.11 drives demand for 802.16a

Scope of 802.16a (2/3) 802.16a is an IEEE Standard for Local and metropolitan area networks (MAN), and specifies an air interface for fixed broadband wireless access systems operating between 2 to 11 GHz. 802.16a defined 3 non-interoperable PHYs : Single Carrier、OFDM and OFDMA. The MAC is TDMA or FDMA. Subscriber Station Base Station

Scope of 802.16a (3/3) System specifications of 802.16a OFDM mode. ETSI (fs/BW=8/7) RF frequency 2-11GHz FFT Size 256 Effective subcarriers 192 Bandwidth (MHz) BW Guard time (us) Tg Data time (us) Tb Symbol time (us) Tg + Tb Subcarrier spacing (kHz) ∆f Sampling rate (MHz) Fs = BW x 8/7 Maximum data rate (Mbps) (BW=28MHz, 64QAM, code rate 3/4) 104.73

Channel Model for Simulation

Channel Model (1/3) Channel profile : SUI-1 SUI-2 SUI-3 SUI-4 SUI-5 Tap1 Tap2 Tap3 K factor SUI-1 Delay (us) 0.4 0.9 3.3 Power (dB) -15 -20 Doppler Frequency (Hz) SUI-2 1.1 1.6 -12 0.2 0.15 0.25 SUI-3 0.5 -5 -10 0.3 SUI-4 1.5 4 -4 -8 SUI-5 10 0.1 2 2.5 SUI-6 14 20 -14

Channel Model (2/3) BW = 1.75MHz (for subcarrier index -127~128 )

Channel Model (3/3)

Transceiver Block Diagram

Transceiver Block Diagram -- Transmitter User Data Scrambler RS Encoder Convolution Encoder Interleaver : Simulink : C++ To DAC IFFT (256-point) Frame Shaping Pilot Insertion QAM Mapper Random Generator Inner Transmitter 802.16a OFDM mode Transmitter Block Diagram

Transceiver Block Diagram -- Receiver Interpolator De- rotator FFT (256-Point) FEQ Slicer To FEC Coarse Symbol Boundary Detection and Fractional Part CFO Acquisition NCO FFT Window Pilot Extraction Long Preamble Extraction From ADC Fine Symbol Boundary Detection and Integer part CFO Acquisition LPF WLS Estimator Channel Estimation Integrator Scaling 802.16a OFDM mode Receiver Block Diagram

Coarse Symbol Boundary Detection and Fractional Part CFO Acquisition

Coarse Symbol Boundary Detection (1/3) Guard Interval Short Preamble Guard Interval Long Preamble PN Sequence period = 64 PN Sequence period = 128 64 64 64 64 128 128 Signal Detection, AGC, …… Since the first several samples are used for Signal detection, AGC, ……, we can not sure how many periods(64 samples) of short preamble can be used for symbol boundary detection. Assume that we can get at least 2 complete periods of short preamble.

Coarse Symbol Boundary Detection (2/3) We can simply use delay correlator to detect the reception of short preamble. From [1], we compute the following equations: Where rn is the received signal, P(d) is the delay correlator of length L (for our case, L=64 ), R(d) is the power sum of L consecutive received samples, M(d) is the delay correlator normalized by R(d). The reason of computing M(d) is that, from [1], we have So we can estimate SNR by computing this equation. (dopt is the optimum position for M(d). ) Normalized CFO

Fine Symbol Boundary Detection and Integer Part CFO Acquisition

Fine Symbol Boundary Detection and Integer Part CFO Acquisition (1/3) For 802.16a, MAX CFO = 10.68GHz * ±4ppm = 85.44KHz ≒12.5*(minimum Subcarrier spacing 6.84KHz) Need Integer part CFO Acquisition For 802.16a, CFO can be derived from the following equation: Subcarrier spacing Sample time Fractional part CFO Integer part CFO If CFO=3.2 ∆f，Phase of P(d) will be 2πx 0.8 = 1.6π= -0.4π (∵tan-1 lies in (-π,π] ) ∴0.5πx (ef + ei ) = -0.4π, we have (ef + ei ) = -0.8 ∆f, so we compensate 0.8 ∆f So the total CFO will be 3.2+0.8=4 ∆f

Fine Symbol Boundary Detection and Integer Part CFO Acquisition (2/3) Therefore, for CFO lies in [-2,2] ∆f, we compensate it to 0 ∆f and for CFO lies in [2,6] ∆f, we compensate it to 4 ∆f, and so on… The following figure illustrates the compensation of fractional CFO : Since for 802.16a, the Maximum CFO can be ±12.5 ∆f, the resulting Integer part CFO can be {-12,-8,-4,0,4,8,12} ∆f We adopt correlator bank with 7 sets of correlator to find the correct integer part CFO.

Fine Symbol Boundary Detection and Integer Part CFO Acquisition (3/3) The procedure of Symbol Boundary Detection and CFO Estimation is： (i) Compute Normalized Delay Correlation M(d) and its moving average with length equals to guard interval. (ii) Find the peak of the moving average, and from the phase of its corresponding delay correlation, we find the Fractional CFO. (iii) When the moving Average drops to half of the peak value，we set the position 128 samples right to the peak position as the Coarse Symbol Boundary (iv) Start finding Fine Symbol Boundary at the position of ±16 samples from Coarse Symbol Boundary. (v) Use Long Preamble Correlator Bank at the searching window. The set with peak occurs indicates the correct Integer part CFO, and the peak position is then the Fine symbol boundary. (vi) To handle the situation that “The first path is not the strongest path”, we use a threshold to find the peak. The threshold is set to be the “half of R(d)”, which is half of the power sum of 64 consecutive received samples.

Simulation Result of Symbol Boundary Detection and CFO Estimation (1/2)

Simulation Result of Symbol Boundary Detection and CFO Estimation (2/2)

Tracking Residual CFO--- WLS Estimator, LPF,NCO De- rotator NCO LPF WLS Estimator

WLS Estimation According to the Spec of 802.16a [2] , there’s only 1 oscillator in the receiver. Therefore, we can adopt the Joint WLSE method [3] to find the residual CFO and TFO. [4] WLSE Block Diagram

Low Pass Filter (1/2) From [5], we adopt the PI control LPF. Its transfer function is We can adjust the values of C1 and C2 to make a trade-off between convergence speed and jitter. The block diagram of LPF is shown below: X D C2 C1 Input Output

Low Pass Filter (2/3) best Without AWGN C1=0.5, C2=0.5 Without AWGN

Low Pass Filter (3/3) Simulation under SUI-3, CFO= -12.5∆f, C1=0.25,C2=0.125, Residual CFO= -0.05∆f LPF Output LPF Output SNR=12dB SNR = 20dB

Pilot Extraction, FEQ and Slicer

Simulation Result Simulation under SUI-3, SNR=25dB, CFO= -0.68352 ∆f, Residual CFO=0.05 ∆f, 16QAM, 500 OFDM Symbol transmitted ( 384,000 data bits ), BER=4.87x10-3 Pilots are modulated with BPSK

Tracking TFO--- Scaling, Integrator, and Interpolator

Scaling CFO to get TFO Since we have only 1 oscillator in receiver, we have If the total estimated CFO is e ,we can get TFO (d) by scaling e, i.e. For 802.16a with ETSI channelization, we have the following 5 cases when TFO is fixed to -8ppm. BW(MHz) Tb(us) ∆f(kHz) CFO (∆f) Case 1 1.75 128 125*(2)-4 10.93632 Case 2 3.5 64 125*(2)-3 5.46816 Case 3 7 32 125*(2)-2 2.73408 Case 4 14 16 125*(2)-1 1.36704 Case 5 28 8 125 0.68352

Interpolator (1/4) We use Farrow Structure piecewise parabolic Interpolator to resample signal.

Interpolator (2/4) If TFO < 0, i.e. Receiver clock period < Transmitter clock period)

Interpolator (3/4) If TFO > 0, i.e. Receiver clock period > Transmitter clock period)

Interpolator (4/4) We can modify our interpolator as shown below : mk TFO (m-1,m0,m1,m2) 0~1 >0 <0 (d1,d2,d3,d4) >1 >0 (d2,d3,d4,d5) <0 (d0,d1,d2,d3) If TFO > 0 and mk Overflows If TFO < 0 and mk Overflows Shift Registers mk not Overflow yet

Encountered Problem… Since we use Farrow structure to model the effect of TFO, and compensate TFO, the imperfect property of Farrow structure become serious especially when mk ≈ 0.5 mk≈ 0.5 mk≈ 0 mk≈1

Conclusion and Future Work Several blocks of 802.16a Transceiver have been introduced. The receiver seems work fine under SUI 1~6 with CFO exists. However, the way we model TFO seems not ideal enough, and we can’t have good performance when TFO exists. The short-term job is to find an appropriate way to model TFO. Up-sampling or Using other kind of interpolator Other jobs including outer transceiver (in C++), other imperfect channel effect, and OFDMA mode……

Thank you for your Attention!

Reference [1]Robust frequency and timing synchronization for OFDM Schmidl, T.M.; Cox, D.C.; Communications, IEEE Transactions on , Volume: 45 Issue: 12 , Dec. 1997 , Page(s): 1613 -1621 [2] IEEE 802.16a draft version 7. [3]Joint weighted least squares estimation of frequency and timing offset for OFDM systems over fading channels Pei-Yun Tsai; Hsin-Yu Kang; Tzi-Dar Chiueh; Vehicular Technology Conference, 2003. VTC 2003-Spring. The 57th IEEE Semiannual , Volume: 4 , April 22-25, 2003 [4]Design and Implementation of an MC-CDMA Baseband Transceiver Hsin-Yu Kang; July , 2003 [5]  Interpolation in Digital Modems---Part II: Implementation and Performance Lars Erup, Floyd M.Garden and RobertA. Harris, IEEE Trans. On Comm.1993