OFDM Based Air Interfaces for Future Satellite Systems PhD Viva Presentation Sundarampillai Janaaththanan Supervisors: Prof. Barry G. Evans Dr. Christos Kasparis CCSR
Outline of Presentation Introduction :-: (Sat. Comm., OFDM) Contributions made in this thesis: New Tone Reservation (TR) based PAPR Reduction Technique Improved Digital Pre-distortion algorithm for Satellite Applications First comparison between OFDM based and HSPA based air interfaces both in the Forward and Return satellite links Jan. 2009 2
Satellite & OFDM Traditional Satellite systems - niche geographical areas and broadcasting applications ( but not successful in capturing the global market) Must follow the terrestrial based standards to capture the markets Several European Union Projects (SATIN, MODIS, MAESTRO) - Investigated S-UMTS systems-based on terrestrial WCDMA air interface Intermediate Module Repeaters (IMR)- to penetrate urban and indoor areas Due to the capacity issue in WCDMA based air interface Terrestrial systems are towards OFDM based air interface better cell capacity Eg. WiMax, 3GPP LTE etc. Satellite systems have recently started considering OFDM (commercial reason) For example : S-DMB systems based on DVB-SH standard (OFDM based standard) We mainly investigate OFDM based air interfaces for bi-directional future satellite systems. 1. commercial reason to capture the market. 2. larger delay spread of the integrated terrestrial /satellite channel. Jan. 2009 3
OFDM (1/3) Multicarrier technology with orthogonally overlapping sub-carriers x n 1 N N 1 k0 X k e j2nk N, n 0, 1, 2,, N 1. Cyclic Prefix is used for: - cope with channel delay spread or ISI. - Low complexity Frequency Domain Equalization at Rx. T n=0 n=1 n=2 n=3 n=4 n=5 n=6 Resulting in high Peak-to-Average Power Ratio (PAPR) => increases with the number of sub-carriers. i.e. high fluctuation in the envelope of the OFDM signal. Where N- number of sub-carriers i sample index Jan. 2009 E[ ]- Expectation 4 operator
The on-board Power amplifiers are inherently non-linear. The Payload Characteristics data from the Alcatel Space ( obtained from IST FP 6 MAESTRO project: S-band TWTA) 0 Satellite Payload (2/3) Non-Linear Region AM-AM Compression Region 50 AM-PM 2 45 40 4 35 OBO (db) 6 8 10 12 Saturation Region Phase (degree) 30 25 20 15 10 5 14 20 15 10 5 0 5 10 IBO (db) 0 20 15 10 5 0 5 10 IBO (db) Jan. 2009 5
Non-linear Distortion (3/3) Satellite systems are Power Limited due to on-board nature- HPA must be operated near saturation region to have higher power efficiency High PAPR + inherently non-linear TWTA = Severe Non-linear Distortion Effect of non-linear distortion 1. In band distortion Bit Error Rate (BER) degradation 2. Out-of-band distortion Adjacent Channel Interference (ACI) Compensation techniques must be transmitter based in order to compensate ACI 1 back-off (at the expense of power efficiency) 2 PAPR Reduction => Tone Reservation Technique 3 Power Amplifier Linearization =>Digital Pre-distortion Technique Jan. 2009 6
Tone Reservation Techniques (1/8) In discrete time-domain: In discrete freq.-domain: Reserved sub-carriers are orthogonal to data carriers C * arg min CB PAPR x[ n] arg min max x arg min x CB 0iN 1 i CB Jan. 2009 7
Existing Algorithms (2/8) 1. Fourier Projection Algorithm (FPA) Based on the Projection onto Convex Sets (POCS) technique Iteratively clips and filters the signal 2. Controlled Clipper Algorithm Based on a peak cancelling kernel signal Amplitude scaled version of the kernel is subtracted from the peaks. - convergence is slow. - take large number of iterations to obtain a sub-optimal solution. 3. Active Set algorithm <= applicable only in real domain Most efficient existing algorithm Minimizes the first peak to second peak and then minimizes both of them to third peak, so on. Uses the same kernel signal as in the Controlled Clipper Algorithm. - convergence is faster - complexity increases with the iteration - becomes very complex for complex baseband signals Therefore, there is a need for an efficient optimization algorithm. Jan. 2009 8
Proposed Algorithm (3/8) We propose a gradient decent algorithm due to its computational efficiency. Optimization function is not smooth approximated to obtain the gradient asymtotically (p-norm approximation) original cost function J x lim p x p Asymptotic gradient of the cost function : J p C x b 1 1 * F b, l xb v In the Frequency Domain Update rule with constant step size C i C i 1 x b 1 1 * F b, l x b v Jan. 2009 9
Proposed Algorithm (4/8) In practice C- must be limited due to the regulatory constraints - Power Spectral Mask Power Spectral Density of OFDM based HyperLAN2 The gradient projection algorithm is used ( Gradient + Projection onto constraint signal set) Projection step: Jan. 2009 10
Positions of Reserved Sub-carriers (Pilots)(5/8)(5/8) Data sub-carriers frequency Reserved sub-carriers Position 1- Uniformly spaced locations Position 2- continuous locations (Edge of the frequency band) Position 3- Randomly spaced locations Jan. 2009 11
Results: Unconstrained Performance (6/8) Pr(PAPR>PAPRo) 10 0 10-1 10-2 10-3 Original OFDM TR-Pos. 1 TR-Pos. 2 TR-Pos. 3 active-set [BRIA04] Better Performance than Active set method PAPR reduction depends on pilot locations Random > Equal Spacing > continuous Possibility of different peak cancelling signal with different locations. 10-4 3 4 5 6 7 8 9 10 11 12 13 PAPRo [db] unconstrained case TR compared with Active set algorithm Jan. 2009 12
Results: PSD-constrained Performance(7/8) Pr(PAPR>PAPRo) 10 0 10-1 10-2 Original OFDM TR-Pos. 1 TR-Pos. 2 TR-Pos. 3 Cosntrained TR-Pos. 1 Constrained TR-Pos. 2 Constrained TR-Pos. 3 Reduction performance is significantly affected in all cases. Reserved tones are frozen at constraint values, therefore, limit the performance. 10-3 10-4 3 4 5 6 7 8 9 10 11 12 13 PAPRo [db] Peak Reduction Tones are constrained to the spectral mask level Jan. 2009 13
Summary- Tone Reservation (8/8) Proposed algorithm Low complexity: Algorithms Real Add. Real Multip. Real Div. Proposed 3NL+2M 4NL+8M - Active set 2GLN(I-1)+8GLN 2GLN+2GLNI 4GLN L-oversampling factor, N- FFT size, M- No. of reserved sub-carriers I- iteration number in active set algorithm, G- Polygon approx ( 4) Improved performance compared to the active set algorithm. Spectral constraints can be applied by simply limiting the reserved sub-carriers Frequency domain iterations. Jan. 2009 14
Digital Pre-distortion Technique(1/6) Most efficient Existing approaches are: 1. Look UP Table (LUT) - Popular Technique 2. Polynomial functions TWTA New problem in satellite :- Remote Adaptation of Pre-distorters Assumptions: The feedback signal from the satellite is only degraded by AWGN noise. Low SNR conditions are more realistic Training Sequence OFD Jan. 2009 M PD MOD GES 15 Noisy Feedback sequence
LUT Pre-distorter (2/6) The Secant algorithm is used for Training the LUT Transmit a ramp-up sequence of L samples for training the LUT Jan. 2009 16
Instability of the Secant algorithm(3/6) Amplitude of PD coefficients Amplitude of PD coefficients 1 0.8 0.6 0.4 0.2 0 0 10 20 30 40 50 60 Table Location SNR= 10dB 1 0.8 0.6 0.4 0.2 SNR= 0 db 0 0 10 20 30 40 50 60 Table Location Adaptation becomes unstable over noisy channel (some LUT entries converge to infinity) Jan. 2009 17
Modified Secant Algorithm(4/6) Based on the recently proposed Modified Newton -Raphson Algorithm x k 1 x r k rx f r1 k f x k ' x k 1 r where f ' x k f x f x k x k x k1 k1 r=1 conventional Secant Algorithm r>1 uses a curvature instead of straight line. Faster convergence than Secant but increased complexity with r. Jan. 2009 18
Modified Secant Algorithm(5/6) 10 0 10-1 BER 10-2 10-3 10-4 10-5 10-6 Mod. Secant SNR= 30, inf db Mod. Secant SNR= 10, 20dB Secant SNR= infdb Secant SNR= 25dB Secant SNR= 20dB Secant SNR= 10dB OFDM AWGN 0 2 4 6 8 10 12 14 16 E b /N 0 [db] When SNR 20dB Modified Secant Algorithm Else ( SNR > 20dB) Classical Secant Algorithm In low SNR region, Modified-Secant is stable compared to conventional secant. Jan. 2009 19
Summary Digital Pre-distortion distortion(6/6) LUT Pre-distorter Secant algorithm Becomes unstable in lower SNR, but stable in higher SNR. Modified Secant algorithm Becomes stable in lower SNR. (approx. <20dB) No improvement at larger SNR. Proposal: an adaptive approach can be used by adapting the r according to the SNR in the feedback channel. Jan. 2009 20
Air Interface Comparisons (1/7) Single user scenario in all cases Comparison is performed based on 3GPP standards. Performance comparison Block Error Rate Vs. Eb/No Jan. 2009 21
Multipath Channel Profiles(2/7) Nor. Power Nor. Power Nor. Power 1 0.5 0 1 0.5 0 1 0.5 channel case 1 10 20 30 40 50 60 70 No of Samples channel case 3 10 20 30 40 50 60 70 No of Samples channel case 5 Satellite only with Line of Sight (LOS) Satellite only with Non Line of Sight (NLOS) Satellite with 3 terrestrial repeaters (urban environment) 0 0 10 20 30 40 50 60 70 No of Samples Multipath channel models from measurements in IST FP6 MAESTRO project. Jan. 2009 22
10 0 OFDM Vs. HSDPA- Simulation Results (3/7) 10-1 BLER 10-2 10-3 10-4 HSDPA CH 1 HSDPA CH 3 HSDPA CH 5 OFDM CH 1 OFDM CH 3 OFDM CH 5 OFDM CH 5 CP=68 0 2 4 6 8 10 12 E b /N 0 [db] Performance of HSDPA and OFDM for different mobile channel profiles, assuming no amplifier distortion and perfect channel estimation 6 Rake Finger receiver in HSPDA, ZF Equalization in OFDM HSDPA performs better in Sat. + IMR channel (case 5) Jan. 2009 23
Return Link- SC-FDMA (4/7) Considered in LTE for Return link. Multi user version of SC-FDE. DFT spreading at the Tx- Low PAPR compared to OFDMA. Based on the sub-carrier mapping Localized SC-FDMA (considered) Distributed SC-FDMA Jan. 2009 24
SC-FDMA Vs. HSUPA Results (5/7) 10 0 10-1 SC-FDMA-Ch.1 SC-FDMA-Ch.3 SC-FDMA-Ch.5 HSUPA-Ch.1 HSUPA-Ch.3 HSUPA-Ch.5 BLER 10-2 10-3 10-4 0 1 2 3 4 5 6 7 8 9 10 Eb/No [db] Performance of HSUPA and SC-FDMA for different mobile channel profiles, assuming no amplifier distortion and perfect channel estimation 6 Fingers Rake Receiver in HSUPA MMSE Equalization in SC-FDMA Jan. 2009 25
SC-FDMA + CAZAC Results (6/7) 10 0 10-1 SC-FDMA Ch.1 (amp) SC-FDMA Ch.3 (amp) SC-FDMA Ch.5 (amp) SC-FDMA Ch.1 (amp+cazac) SC-FDMA Ch.3 (amp+cazac) SC-FDMA Ch.5 (amp+cazac) BLER 10-2 10-3 10-4 4 5 6 7 8 9 10 11 12 13 14 E b /N 0 [db] Performance of SC-FDMA for different mobile channel profiles with realistic channel estimation using CAZAC sequence Constant Amplitude Zero Auto-Correlation (CAZAC) sequence based channel Estimation provides improved performance. NOTE: other results are based on the best random pilot sequence obtained via Jan. 2009 trial-and-error. 26
Summary -Air Interface Comparisons (7/7) Forward Link HSDPA performs better in terms of BLER Vs. Eb/No performance - Rake receiver utilizes larger window to detect all the delayed signals. Return Link HSUPA performs better in terms of BLER Vs. Eb/No performance CAZAC sequence provides improved performance in the non-linear channel. The results are not necessarily surprising. (multicarrier air interfaces do not necessarily perform best) More work needs to be done to provide final conclusion. ( Throughput, Spectral Efficiency) Jan. 2009 27
List of Publications Journal Paper: 1. Janaaththanan S., Kasparis C., Evans B.G., A Performance Comparison Study between HSPA based and OFDM based Air Interfaces for Future Satellite, to be submitted to Internal Journal on Satellite and Networking. Feb. 2009. Conference Papers: 2 Janaaththanan S., Kasparis C., Evans B.G., Performance Comparison between Adaptive LUT and Polynomial based Pre-distorters in the Forward Link of Satellite Systems, submitted to 27 th AIAA International Satellite Systems Conference (ICSSC). 3 Janaaththanan S., Kasparis C., Evans B.G., Improved Adaptation Algorithm for LUT-based Pre-distorter with Noisy Training Phase in OFDM based Satellite System, 26 th AIAA International Satellite Systems Conference (ICSSC), San Diego, California, U.S.A, 11-14 June 2008. 4 Janaaththanan S., Kasparis C., Evans B.G., A Gradient Based Algorithm for PAPR Reduction of OFDM using Tone Reservation Technique, Vehicular Technology Conference, 2008. VTC Spring 2008. IEEE, pp.2977-2980, 11-14 May 2008. 5 Janaaththanan S., Kasparis C., Evans B.G., Comparison of SC-FDMA and HSUPA in the Return Link of Evolved S- UMTS, International Workshop on Satellite and Space (IWSSC), pp.56-60, Austria, 13-14 Sept. 2007. 6 Janaaththanan S., Kasparis C., Evans B.G., Feasibility Study of Adaptive LUT-based Pre-distorter for OFDM in Non-linear Satellite Downlink Channel, International Workshop on Satellite and Space (IWSSC), pp.126-129, Madrid, Sept. 2006. 7 Janaaththanan S., Kasparis C., Evans B.G., A Comparison between OFDM and SC-FDE over Wideband Satellite Downlinks, 24 th AIAA International Satellite Systems Conference (ICSSC), San Diego, California, U.S.A, 11-14 June 2006. Jan. 2009 28
Thank You Any Questions?? Jan. 2009 29