May 2003 doc.: IEEE 802.15-03/141r3 November, 2004 Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs) Submission Title: [Multi-Band OFDM Interference on In-Band QPSK Receivers Revisited] Date Submitted: [14 November, 2004] Source: [Celestino A. Corral, Shahriar Emami, Gregg Rasor] Company [Freescale] Address [3301 Quantum Blvd., Boynton Beach, Florida, USA 33426] Voice:[561-739-3280], FAX: [ ] Re: [] Abstract: [This document provides simulation and theoretical results that demonstrate MB-OFDM is an extremely harmful type of interference to wideband in-band QPSK systems such as C-band TVRO receivers.] Purpose: [For discussion by IEEE 802.15 TG3a.] Notice: This document has been prepared to assist the IEEE P802.15. 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 P802.15. Celestino A. Corral et al., Freescale Jai Balakrishnan et al., Texas Instruments

Multi-band OFDM Interference on In-Band QPSK Receivers Revisited November, 2004 Multi-band OFDM Interference on In-Band QPSK Receivers Revisited Celestino A. Corral, Shahriar Emami and Gregg Rasor Freescale Semiconductor 3301 Quantum Blvd. Boynton Beach, Florida November 14, 2004 Celestino A. Corral et al., Freescale

November, 2004 Motivation Goal: To provide additional simulation results for the source of interference in MB-OFDM modulation. Focus is on interference to in-band high data rate wireless systems, particularly TVRO satellite receivers using QPSK modulation. Note: Multi-band UWB, including MB-OFDM, concentrates its energy in a narrower bandwidth than a comparable DS-UWB system under equal effective isotropic radiated power (EIRP). The filter captured energy is higher. Approach: Analyze the source of interference from a time and spectrum perspective. Additionally: Clarify initial results of Portland meeting. Celestino A. Corral et al., Freescale

Multi-band UWB Power Recap November, 2004 Multi-band UWB Power FCC states power spectral density for UWB devices must be -41.3 dBm/MHz in band between 3.1 and 10.6 GHz. Since multi-band signals hop over a selected band of frequencies, the power spectrum is scaled by the hop and averaged over the band. The resulting power spectral density is made equal to a system over any arbitrary band. Multi-band spectrum PSD level f1 fx f2 Integrate the spectrum over band and average by band To implement equal PSD over hop bandwidth, we need Recap requiring a power scaling. Celestino A. Corral et al., Freescale

Multi-band UWB Power Recap November, 2004 Equate power Both systems have equal range and total equal power. Actual MB-OFDM PSD over its transmission bandwidth. Assuming DS-UWB bandwith is 2 GHz and MB-OFDM bandwidth is 528 MHz. Celestino A. Corral et al., Freescale

OFDM and AWGN Subcarriers are orthogonally spaced in frequency. November, 2004 OFDM and AWGN Subcarriers are orthogonally spaced in frequency. Data modulation on subcarriers randomizes amplitude and phase. Peak-to-average approaches that of AWGN as the number of subcarriers increases, but is bound to 10 log (N). Peak-to-Average Power Plots f1 f2 f3 f4 … number of subcarriers Some similarities are evident Celestino A. Corral et al., Freescale

November, 2004 OFDM and AWGN Temporal Snapshot PDF AWGN Both signals have the same average power and identical PDF… OFDM But they’re not the same! Celestino A. Corral et al., Freescale

OFDM and AWGN Energy in time equals energy in spectrum November, 2004 OFDM and AWGN Energy in time equals energy in spectrum Spectral densities are inversely proportional to the bandwidth of the signal. OFDM concentrates more of its energy over a narrower spectrum than DS-UWB, hence higher spectral density. This is evident at the output of the matched filter with optimum sampling. In-band filter bandwidth 0.528 Spectral densities MB-OFDM spectrum DS-UWB spectrum Amplitude 3.1 5.1 f (GHz) AWGN OFDM Celestino A. Corral et al., Freescale

OFDM and AWGN Matched Spectral Densities November, 2004 AWGN OFDM If the power spectral densities are equal, OFDM will have less energy than DS-UWB. Another viewpoint: At a given spectral density for OFDM, DS-UWB can transmit more energy! Celestino A. Corral et al., Freescale

Ungated OFDM BER Results November, 2004 Ungated OFDM BER Results Higher Spectral Density Results in Error OFDM DS-UWB Ungated OFDM with equal EIRP is more harmful interference than DS-UWB DS-UWB spreads its energy over greater bandwidth, so it produces less interference Celestino A. Corral et al., Freescale

OFDM Modeled as Gated AWGN November, 2004 OFDM Modeled as Gated AWGN In doc. 315r0 the MB-OFDM results were with two phenomena captured: PSD growth due to equal EIRP Additional interference due to averaging of EIRP over the hop depth. We need to equate the PSD so that the averaging of the EIRP produces the actual PSD growth (i.e., we need to make the PSD’s of each interference the same). 3 hops AWGN 9 dB Celestino A. Corral et al., Freescale

interference not present November, 2004 Gated AWGN Revisited Symbol Error Rate (QPSK): Bit Error Rate: interference present Interference is Gated: interference silent New Bit Error Rate: = 0 interference present interference not present Celestino A. Corral et al., Freescale

Consider Interference-to-Noise November, 2004 Consider Interference-to-Noise Probability of Bit Error: where Interference-to-Noise Ratio Asymptotic Behavior (Ns = 0): Probability of bit error as time of interference presence increases (gating approaches continuous operation) Asymptotic Loss of Gated Noise Model Relative to Continuous Noise: Celestino A. Corral et al., Freescale

November, 2004 BER versus INR for 3 Hops Lower INR results in less interference, but not zero. In evaluating INR we cannot assume users are cognizant of regulatory rules. DS-UWB causes lower interference relative to MB-OFDM when latter is modeled as gated noise. Celestino A. Corral et al., Freescale

Plot of Theoretical Loss for Gated Noise Source November, 2004 Plot of Theoretical Loss for Gated Noise Source Evaluating: Lower INR results in less loss (back-off), but not zero. Loss is higher for longer hops DS-UWB is always lower interference relative to an MB-OFDM system. Celestino A. Corral et al., Freescale

Filtered MB-OFDM Revisited November, 2004 Filtered MB-OFDM Revisited For filtered MB-OFDM, it is assumed that the filter rise time is still sufficient to capture the full interference levels. Filtering consists of the ideal rejection of subcarriers outside the desired bandwidth. Energy is made equal over the bandwidth of the filter by scaling the interference using 10 log(M/N) where M is the number of subcarriers captured and N is total number of subcarriers. Variance: Celestino A. Corral et al., Freescale

Filtered MB-OFDM Results November, 2004 Filtered MB-OFDM Results Ideal filtering implemented: 40 MHz bandwidth corresponds to 8 subcarriers passed, all others infinitely rejected. Power scaled so that PSD of MB-OFDM and AWGN are the same. As Eb/No increases, trend seems to be that SER improves. Celestino A. Corral et al., Freescale

Cipped MB-OFDM Results November, 2004 Cipped MB-OFDM Results Clipping level set at 9 dB per the MB-OFDM proposal. Clipping has no impact on BER results. Impulsive characteristic is suppressed, but main contributor is still the bursty nature of the MB-OFDM interference. Celestino A. Corral et al., Freescale

Gated Noise Interference with FEC November, 2004 Gated Noise Interference with FEC Convolutional code, constraint length K = 7 with hard decision, yields about 5 dB coding gain for all cases. No interleaving performed. FEC improves BER performance of all interference. MB-OFDM as gated noise is still worse interferer. Celestino A. Corral et al., Freescale

Impulse Radio Comparisons November, 2004 Impulse Radio Comparisons PRF = 22.2 MHz Impulse radio modeled as gated AWGN process similar to MB-OFDM. Pulse width is 2 nsec, corresponding to 500 MHz bandwidth. EIRP averaged over the hop depth of the gated noise model for MB-OFDM. Practical PRF range considered. PRF = 2.22 MHz Celestino A. Corral et al., Freescale

Impulse Radio Comparisons November, 2004 Impulse Radio Comparisons For very high PRF, impulse radio approaches AWGN. For lower PRF, SER for impulse radio rises moderately. Under constraint of identical 500 MHz bandwidth, impulse radio interference is lower than MB-OFDM modeled by same gated noise process. Celestino A. Corral et al., Freescale

November, 2004 Conclusions Ungated OFDM is a more harmful interferer than DS-UWB under equal EIRP constraint because the energy is concentrated over a narrower bandwidth. Gated noise model was used to evaluate MB-OFDM interference under equal PSD constraint. Results show higher interference from gated noise than continuous noise. Gated noise model was extended to handle interference-to-noise ratios and theoretical loss difference between systems established for lowest hop depth N = 3. Celestino A. Corral et al., Freescale

November, 2004 Conclusions Filtered MB-OFDM seems to indicate that narrower filtering improves SER performance slightly. However, results are optimistic as they account for “ideal” filtering. Results for clipped MB-OFDM show basically no difference when compared to unclipped MB-OFDM. All interference sources benefit from FEC, but MB-OFDM is still worse than DS-UWB. Impulse radio interference is less than that of MB-OFDM when both are modeled as gated AWGN processes with equal 500 MHz bandwidths and over practical PRF ranges. Celestino A. Corral et al., Freescale

Clarification of Results Presented in Doc 412r0 – APD Analysis November, 2004 Clarification of Results Presented in Doc 412r0 – APD Analysis APD is a methodology that captures only amplitude info: Amplitude (A) in dB as ordinate, 1-CDF(A) plotted as abscissa. Slide 3 clearly states “For full impact assessment, knowledge of the victim system’s modulation scheme and FEC performance is needed.” In other words, APD is only a piece of the puzzle. APD has value, but results must be considered under the basis of the method’s limitations. Specifically, amplitude data alone is not sufficient, greater scrutiny is needed. We provide examples of waveforms with similar APDs and different interference potential. Celestino A. Corral et al., Freescale

Three Different Signals November, 2004 Three Different Signals AM modulated signals: Sinusoid Quasi-Sinusoid Scrambled Sinusoid Which Waveform Interferes More? Celestino A. Corral et al., Freescale

APD Results APDs Are The Same! November, 2004 APD’s treat only envelope of waveforms. Celestino A. Corral et al., Freescale

Different Spectra November, 2004 Sample Signals Detail of Scrambled Sinusoid The interference potential of signals cannot be determined by APD analysis in isolation. Victim bandwidth, center frequency, modulation, etc. play a role. More information is needed! APD analysis especially breaks down when considering the impact of FEC. Celestino A. Corral et al., Freescale

PDF of Signals PDF in Slide 25 of Doc 412 Actual PDF November, 2004 var = 0.5 var = 2 Even with finite values, peak signal is higher! This PDF shows Gaussian noise and OFDM have the same variance (power). But this is not the case: MB-OFDM has 6 dB more power. PDF cannot be “averaged” as signal. This gives the impression OFDM is more benign than AWGN, which it is not. This PDF clearly shows approximately 6 dB greater power (4X variance) of OFDM. This is at output of matched filter at optimum sampling point. This PDF is present at a duty cycle of 26%; but it is not “averaged.” For the other cases, variance = 0. Celestino A. Corral et al., Freescale

Interference Conditions November, 2004 Interference Conditions Slides 27—29 confirm results for simplified case of only gated noise interference present (i.e., no noise). Considers more “realistic” case of noise always present. Analysis then considers Eb/(No + Io) with receiver at some fixed Eb/No; increase Io after that. By judicious selection of No, impact of Io can be suppressed; this is not representative of interference effects, only noise effects! Analysis presented here for slides 14—16 are representative of Eb/(No + Io) effects under high SNR, which is case for TVRO systems. Celestino A. Corral et al., Freescale

Back-Up Material: OFDM Correlation November, 2004 Back-Up Material: OFDM Correlation OFDM is additive noise. Compared autocorrelation of OFDM and AWGN processes. OFDM exhibits significant autocorrelation compared to AWGN. Celestino A. Corral et al., Freescale

Back-Up Material: OFDM Correlation November, 2004 Back-Up Material: OFDM Correlation Compared two different OFDM systems: 128 (528 MHz) 256 (1.056 GHz) Autocorrelation improves as more subcarriers (and corresponding wider bandwidth) are employed. Celestino A. Corral et al., Freescale

Correlation Effects OFDM signal is highly correlated; it is not white. November, 2004 Correlation Effects OFDM signal is highly correlated; it is not white. Autocorrelation improves with more subcarriers and larger bandwidth. OFDM is additive noise and approaches Gaussian with large number of subcarriers. Receivers are typically designed for AWGN. Receivers expect to operate on uncorrelated noise samples. For OFDM interference, receiver performance will be inferior to AWGN. Celestino A. Corral et al., Freescale