Sept 2004 doc IEEE 802 15 040412 r

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Project: IEEE P 802. 15 Working Group for Wireless Personal Area Networks (WPANs) Submission Title: [In-band Interference Properties of MB-OFDM] Date Submitted: [9 Sept, 2004] Source: [Charles Razzell] Company [Philips] Address [1109, Mc. Kay Drive, San Jose, CA 95131, USA] Voice: [+1 408 474 7243], FAX: [+1 408 474 5343], E-Mail: [charles. razzell@philips. com] Re: [Extension of previous APD analysis in 802. 15 -04/326 r 0 and address points raised in 315 r 0 ] Abstract: [Presents in-band interference properties of MB-OFDM as revealed by statistical properties (APDs) and by impact to BER curves for a QPSK transmission system] Purpose: [To correct potential misapprehensions concerning the interference impact of MB-OFDM. ] Notice: This document has been prepared to assist the IEEE P 802. 15. It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor acknowledges and accepts that this contribution becomes the property of IEEE and may be made publicly available by P 802. 15. Submission 1 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Part 1 APD Plots and their Implications for MB -OFDM Submission 2 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Amplitude Probability Distributions • APD methodology is favored by the NTIA in assessing interference impact of UWB waveforms • For non-Gaussian interference, APD plots provide helpful insight into potential impact on victim receivers. • For full impact assessment, knowledge of the victim system’s modulation scheme and FEC performance is needed Submission 3 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Example APD plot (for Rayleigh Distribution) Amplitude (A) in d. B is plotted as the Ordinate 1 -CDF(A) is plotted as the Abscissa Plotting the natural log of the probabilities on a log scale provides scaling similar to Rayleigh graph paper. P(A>10 d. B) = exp(-10) = 4. 54 x 10 -5 ; Submission P(A>-30 d. B) = exp(-0. 001) = 0. 999 4 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 APD plots for continuous OFDM signals as number of QPSK sub-carriers is varied As the number of sub-carriers used increases, the approximation to the Rayleigh APD plot improves. This can be expected due to the Central Limit Theorem. Submission 5 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 APD plots for continuous OFDM with 128 sub-carriers as receiver bandwidth is varied Using receiver filters of increasing bandwidths yields a similar result: approximation to Rayleigh APD is good for b/w 20 MHz Submission 6 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Analytic Expression for APD of OFDM waveforms We have seen that for measurement bandwidths of 20 MHz, the APD of OFDM closely approximates that of a Rayleigh distribution. This can be expected because the in-phase and quadrature components will both tend towards a Gaussian distribution due to the central limit theorem. Assuming this approximation to be perfect, we can write a closed form expression for the APD of OFDM Submission 7 C. Razzell, Philips

Sept 2004 Submission doc. : IEEE 802. 15 -04/0412 r 0 8 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Analytically Derived APD Plot for MBOFDM % APD plots d = 3*165/128; % duty cycle x=linspace(-20, 15); rsq=10. ^(x/10); apd 3=-rsq/d - log(d); apd=-rsq; semilogx(apd 3, x, apd, x) xlabel('ln(P(A>ordinate))') ylabel('Amplitude [d. B]') legend('MB-OFDM', 'cont. OFDM') axis([-10 -0. 01 -20 15]) grid Submission 9 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulated APD plots for continuous and 3 -band OFDM, using 128 sub-carriers Signal/interferer is normalized to unit power 0 d. BW. 1. 8 % Submission Probability of noise amplitude exceeding signal amplitude is given by abscissa value at the intersection of a horizontal SIR line with the APD curve. 10 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulated APD for MB-OFDM as a function of victim Rx bandwidth Victim Rx bandwidth has a significant impact on the APD plots: generally speaking, lower receiver bandwidths “experience” a more benign version of the APD. Submission 11 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulated APD for 1 MHz PRF Impulse as a function of victim Rx bandwidth APD plots for this 1 MHz PRF impulse show significantly higher peaks for large receiver bandwidths {20, 50 MHz}. At lower received bandwidths, APD plots are strikingly similar to those for MB-OFDM (Flipping between this and the previous slide may help illustrate this point. ) Submission 12 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Peak Received Powers As a Function of Receiver Bandwidth The impulse radio’s peak power consistently scales with 20 log(BW). MB-OFDM advantage The continuous OFDM signal (ofdm 1) has a peak power that scales with 10 log(BW) The 3 -band OFDM signal looks like a “hybrid” signal. For lower Rx bandwidths its peak power tracks with the 1 MHz impulse radio, but at 10 MHz and above the slope reverts to that of pure OFDM. Submission 13 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulated APD Curves for OFDM and Impulse Radios in 50 MHz bandwidth 10 MHz PRF impulse radio has nearly identical APD to 1/3 duty cycle OFDM in region of interest. 3 MHz and 1 MHz PRF radios have significantly higher SIR ratios corresponding to the 1. 8% P(A>ord. ) line than the 3 band OFDM system. Submission 1. 8 % All these impulse radios would be permitted under current part 15 f legislation. 14 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Single dominant source of interference may not reflect real scenarios… • All the above APD analysis has assumed that the dominant source of interference is a single instance of the considered waveform • For this to be true: – A single interferer must be very close to the victim receiver such that it can overwhelm: • The thermal noise of the receiver • The additive combination of other uncoordinated UWB and other interferers • Examples of aggregate (Noise + Interference) APDs follow Submission 15 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 APD plots of 1/3 duty cycle OFDM combined with thermal receiver noise Submission 16 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 APD Conclusions • Using the NTIA APD methodology for the worst-case scenario of a single dominant interferer shows: – That the required SIRs for low PRF impulse radios are greater than those needed for the 3 -band OFDM waveform for cases where the victim receiver band exceeds the impulse PRF by a factor of 5 (or more). – The APD plots for lower bandwidth victim receivers show that peaks of the MB-OFDM signal are significantly attenuated by the Rx filter, bringing them closer to the ideal Rayeligh APD. – That peak interference powers due to MB-OFDM are similar to those caused by a 1 MHz PRF impulse radio for <10 MHz victim receiver bandwidths, whereas for >10 MHz receiver bandwidths, significantly lower peak powers are obtained for MB-OFDM. • Receiver thermal noise and other external interference sources will have a mitigating effect on the APD of an interfering MB-OFDM signal Submission 17 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Part 2 MB-OFDM Interference Impact to Inband QPSK transmissions Submission 18 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Background • Document 802. 15 -04/315 r 0 showed large ( 9 d. B) increases in required S/I ratios required when MB-OFDM was the sole source of unwanted interference • These results seemed intuitively unreasonable and therefore merited further investigation • Uncoded QPSK transmissions of circa 33 MHz bandwidth (66 Mbps) were used as basis for comparison Submission 19 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 QPSK Transmission System BIT GENERATOR ERROR COUNTER Submission MULTIPLEXER DEMULTIPLEXER SYMBOL MAPPER HARD DECISIONS 20 16 x UPSAMPLE BY ZERO INSERTION RRC Filter with 33 MHz 3 d. B bandwidth OFDM INTERFERENCE GENERATOR (OR AWGN) + DECIMATION RRC Filter with 33 MHz 3 d. B bandwidth C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Interference Scenario 33 MHz (8 sub-carriers) Each OFDM sub -carrier is modulated with random QPSK symbols QPSK System operates within this bandwidth. The bandwidth is defined by a RRC filter with =0. 5 Submission 21 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 33 MHz QPSK System with AWGN Submission 22 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 33 MHz QPSK System with Continuous OFDM Submission 23 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Continuous OFDM signal causes fewer errors than WGN for same S/(I+N) • This claim may seem counter-intuitive at first • Consider that at high SNRs, errors are caused by the tails of the Gaussian distribution (see “Error Region”, next slide) • But with only 8 relevant sub-carriers the OFDM waveform is limited to 256 states in each of I and Q dimensions – Tails of the distribution poorly approximate Gaussian noise. Submission 24 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Monte Carlo Simulated PDFs of received symbols conditioned on txbits=‘ 1, 1, 1, …’ Eb/Io=7 d. B 500, 000 transmitted bits Probability Density ERROR REGION P(error) = area under the curve Submission Real(rxsymbol) [V] 25 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Output states of 8 -point IFFT with all 65536 possible QPSK symbol sets Amplitude is Bounded over all possible QPSK symbol permutations Filter memory will add more states, but tails of distribution will remain limited in amplitude Submission 26 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Prediction for ¼ duty cycle noise bursts • Combined impact of 3 -band hopping, zero prefix and guard interval is: 165*3/128 = 3. 8672 – We will approximate the duty cycle ratio d = 4 • During, zero noise power periods zero bit errors should occur – Average BER is reduced by a factor of d • During active noise bursts, noise power is d times higher than the long term average – Corresponding SNR reduced by a factor of d Submission 27 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulation with ¼ duty cycle noise bursts as interferer Expected reference for ¼ duty noise bursts Previous Reference for uncoded QPSK Submission 28 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulation with ¼ duty cycle OFDM as interferer Expected reference for ¼ duty noise bursts Previous Reference for uncoded QPSK Submission 29 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 How meaningful is ¼ duty-cycle noise/interference? • The above plots assume that for ¾ of the time, the system noise temperature is 0 Kelvin. – We want to be more realistic than that • Let’s assume the QPSK victim has a constant Eb/No of 10 d. B (the uncoded BER is expected to be erfc(100. 5)/2 3. 87 x 10 -6). • Vary Eb/(No+Io) by introducing ¼ duty cycle MB-OFDM, starting with Io=0 Watts and increasing Submission 30 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Simulation with ¼ duty cycle OFDM + Continuous AWGN <2 d. B Submission 31 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 QPSK BER Conclusions • A continuous OFDM interferer has a more benign error inducing property than AWGN when each is applied at the same S/(I+N) • Under conditions of zero thermal noise, where the interferer has a fixed duty cycle, d, the average BER is closely bounded by • Realistic conditions call for a non-zero value for background thermal noise – In a reasonable test case, deviation of the BER curve from the AWGN case was limited to 2 d. B Submission 32 C. Razzell, Philips

Sept 2004 doc. : IEEE 802. 15 -04/0412 r 0 Overall Conclusions • Impulse radios showed a more harmful APD plot than 3 -band MB-OFDM for all cases where (Rx Bandwidth)/PRF 5. • Low bandwidth ( 5 MHz) cases have also been simulated, revealing close resemblance of the APDs to impulse radios of the same PRF, and much lower peak-to-mean ratios compared to the wideband case. • Testing the impact of MB-OFDM on a QPSK transmission system showed that the required SNR increase is always less than 10 log(d), but in realistic scenarios, with continuous AWGN also present, the impact was reduced to below 2 d. B. Submission 33 C. Razzell, Philips