January 2003 Joe Kwak InterDigital Communications Corporation IEEE 802.11-03/100r1 doc: IEEE 802.11-03/898r0 November 2003 PSNI: New PHY Measurement for Link Quality comparative measurements of receiver output quality to support network management Joe Kwak InterDigital Communications Corporation Submission Joe Kwak, InterDigital

Outline Need for new PHY measurements January 2003 IEEE 802.11-03/100r1 Outline Need for new PHY measurements RCPI and PSNI Relation to SNR in Demodulator Perceived Signal-to-Noise-plus-interference Indicator (PSNI) Definition PSNI Analysis: Relation to EbNo, SNR and BER Use of variance to align fading channels with AWGN Required measurement sample size PSNI as “black box” performance specification Motions to incorporate PSNI Joe Kwak, InterDigital Submission

Need for New PHY Measurements RSSI is defined at antenna input connector but is not fully specified: no unit definitions, no performance requirements (accuracy, testability). Since so little about RSSI is specified, it must be assumed that widely variant implementations already exist. It is not possible to compare RSSIs from different STAs and perhaps not even from different channels/PHYs within same STA. RSSI may have limited use for evaluating AP options within a STA and within a given PHY, but not between PHYs. RSSI is rescaled between DSSS and OFDM PHYs. RSSI is clearly not useable by network management for handoff or load balancing. RSSI from one STA does not relate to RSSI from any other STA. In high interference environments, RSSI is not an adequate indicator of desired signal quality, since it indicates the sum of desired signal + noise + interference powers. Proposed RCPI provides quantized, objective input power measure (S+N+I). Proposed PSNI provides quantized, comparative measure of received signal quality [observed S/(N+I)] for all channels/rates and among all PHYs and between all STAs. Joe Kwak, InterDigital Submission

PHY Measurement Architecture RCPI measures total RF Power at antenna input connector A. PSNI measures observed S/(N+I) within demodulator but normalizes measurement for FER at E. AGC B C D Demodulator and tracking loops (PHY specific) E Radio front end FEC Decoder (optional) Frame Check (CRC) A/D A A: Total RF power, RF S/(N+I) from each AP C&D: Bit Error Rate (BER) @each data rate from each AP B: BB S/(N+I) from each AP (BB power constant by AGC) E: Frame Error Rate (FER) @each data rate from each AP Joe Kwak, InterDigital Submission

Measure PHY Demod Input (power) and Output (QOS) Accurate S/(N+I) measurement at A is interesting but because RF/demod implementations vary widely, it cannot be used comparatively between STAs to evaluate delivered signal quality. Accurate FER measurement at E is ideal quality measure, but cannot be measured frame by frame. FER can only be accurately measured over 100s-1000s of frames. Also, FERs are comparable only at same frame size and data rate. -80dbm Good STA 10E-5 -80dbm Good STA 10E-5 A (dBm) E (FER) A (dBm) E (FER) Med STA 10E-4 Med STA 10E-5 -78dbm -80dbm Marginal STA Marginal STA 10E-2 -75dbm 10E-5 -80dbm Signal at same objective SNR Signal at same subjective SNR Measure RCPI power at A. Measure PSNI quality in middle, but specify PSNI with FER at E. Joe Kwak, InterDigital Submission

RCPI and PSNI Relation to SNR in Demodulator Received Channel Power Level dBm (S + N + I) Operating Margin Required Min RSPL Level Desired Signal Power dBm Interference Power at Input Boltzman’s C (-198dBm/Hz/K) Antenna Connector: Input Power Level (S+ N + I) Input Analog SNIR Ratio Thermal Input Noise Level (-100dBm) Theoretical SNR for required BER (RCPI at antenna connector) NBW = 22MHz = 73.4dB Temp = 290K = 24.6dB Channel Impairments (CI) (fading + multipath + etc, = 0 in AWGN) dBm Total Modem Implementation Losses (TML) FEC Decoder Loss, if any Demodulator Loss Rx Amp Noise Figure + IM Distortion Observed Digital (PSNI in demodulator) Input SNR Ratio Total Channel Condition Losses o Joe Kwak, InterDigital Submission

PSNI: Demodulator-specific, Post-processing Estimator of Observed S/(N+I) and BER/FER. All digital demodulators use tracking loops and complex post-processing to demodulate received symbols. Many internal demodulator metrics are proportional to perceived S/(N+I). Examples: PSK: baseband phase jitter and received Error Vector Magnitude (EVM) DSSS: spreading code correlation quality OFDM: frequency tracking and channel tracking stability OFDM: EVM on pilot channels (cf. Johnson, 802.11-03/844r0) All FEC modes: corrected bit rate in FEC decoder All modes: EVM on data symbols (cf. Kwak, 802.11-03/773r2) Demodulator internal metrics are available on a frame-by-frame basis. Demodulator metrics proportional to S/(N+I) are available at all data rates. Demodulator internal metrics may be calibrated with respect to actual FER performance to accurately indicate output signal quality in AWGN, fading and with interference degradations. Such demodulator internal metrics are fast estimators of S/(N+I) in both interference environments and interference-free (noise only) environments. TGK NEED NOT specify which demodulator metrics to use, but needs only to specify how the quantized PSNI indicator relates to S/(N+I) and FER. Joe Kwak, InterDigital Submission

PSNI Concept: Measure Output Signal Quality Specified like RSSI: 8-bit unsigned value, monotonically increasing with increasing S/(N+I). PSNI shall be logarithmically scaled to perceived S/(N+I) which relates directly to FER performance. Specify single PSNI output value for each data rate using FER point: first point to “anchor” indicator, additional points to quantize and scale indicator slope and range of values. PSNI output values reflect output FER and are specified in AWGN and in one representative fading channel. Specify accuracy of PSNI in AWGN to be +/- 2.0dB in AWGN and +/- 4.0dB in fading channels. PSNI range shall span the lower 43 dB portion of the operating range of S/(N+I) to cover high FERs at data rates from 1 to 54 Mbps. Joe Kwak, InterDigital Submission

PSNI specified on BER/FER curves Joe Kwak, InterDigital Submission

PSNI Normative Specification Text The PSNI indicator is a measure of the perceived, post-processing signal-to-noise-plus-interference (S/(N+I)) ratio in the demodulator. The allowed values for the Perceived Signal to Noise Indicator (PSNI) parameter shall be an 8 bit value in the range from 0 through 255. This parameter shall be a measure by the PHY sublayer of the perceived signal link quality observed after RF downconversion and is derived from internal digital signal processing metrics of the demodulator used to receive frames on that link. PSNI shall be measured over the PLCP preamble and over the entire received frame. PSNI is intended to be used in a relative manner, and it shall be a monotonically increasing, logarithmic function of the observed link S/(N+I). Specified PSNI performance shall be measured over no less than 1000 PPDUs from the same transmitter. PSNI accuracy and range shall be specified in AWGN and fading for each data rate as follows: Theoretical FEC coding gain assumed in FER calculations: R = 1/2, 5.4dB gain R = 2/3, 4.7dB gain R = 3/4, 4.4dB gain PSNI SPECIAL VALUE: “0” shall indicate inability to measure PSNI When PSNI exceeds high end of measurable range for a given data rate, maximum PSNI for that rate shall be reported. Fading channel model is IEEE exponential ray decay with 50nsec decay time. Joe Kwak, InterDigital Submission

Data Rate/Modulation Adjustments (DRMx) Used to Offset BER Curves Table 1: DRM Rate/Modulation Adjustments Joe Kwak, InterDigital Submission

PSNI = 76 is equivalent to Observed OSNIR = 3.7dB Example: PSNI = 76 = OSNIR BERs vary based on FEC coding used at each data rate. FERs vary based on BER and PPDU length. Note: in any STA, PSNI will vary only as a result of changing Channel Conditions or changing received Desired Signal Power Level. Single STA switching data rates will report same PSNI at either rate. Note: for efficiency, all STAs should operate at highest data rate possible while maintaining acceptable FER (QOS). Any STAs at these op points will report PSNI = 76 PSNI = 76 is equivalent to Observed OSNIR = 3.7dB Joe Kwak, InterDigital Submission

PSNI Analysis: Relation to Observed Eb/No PSNI = 0 is selected for a post-processing, Observed Eb/No (OEbNo) equal to 4.4dB, for BPSK at 1Mbit/s data rate. 6 units (steps) per dB is selected to provide 43 dB range in 8 bit PSNI value. So for 1 Mbit/s BPSK operation, PSNI = 6*[OEbNo - 4.4dB]. In general for all other data rates and modulations, PSNI = 6*[OEbNo - 4.4dB + DRMx - CFy] , where DRMx is an S/N adjustment unique for each data rate/demodulation combination. DRMx values are calculated in Table 1 , as shown on page 11, and where CFy is a hardware-specific factor used to account for implementation variances in each FEC decoder in the STA. CFy = CGtheo - CGact = actual FEC decoder loss, for each decoder at each specified FER point. CGtheo values are listed on page 11. When no FEC decoder is used CFy = 0. This relation is the foundation of the PSNI measurement. . Joe Kwak, InterDigital Submission

PSNI Analysis: Relation to Input SNIR (ISNIR) SNR = C / N, where Eb = C * Tb, N = No * NBW (noise BW) and DR = 1 / Tb So SNR = -------------- = -------------- = EbNo * DR / NBW In db: SNR = EbNo + DR - NBW , where EbNo is shorthand for Eb/No in dB. For DR = 1 Mbit/s and NBW + = 22 MHz, SNR = EbNo + 60dB - 73.4dB = EbNo -13.4dB In general, SNR = EbNo -13.4dB + DRMx, with DRMx from Table 1. and so EbNo = SNR + 13.4dB - DRMx, and OEbNo = OSNIR + 13.4dB - DRMx From page 15 we have: PSNI = 6*[OEbNo - 4.4dB + DRMx - CFy], and substituting for OEbNo, PSNI = 6*[(OSNIR + 13.4dB - DRMx) - 4.4dB + DRMx - CFy], and PSNI = 6*[OSNIR + 9.0dB - CFy] Since ISNIR = OSNIR + TML + CI, where TML is the modem implementation loss and CI is the sum of all channel impairments, we have PSNI = 6*[(ISNIR-TML-CI) + 9.0dB - CFy] Eb / Tb Eb * DR No * NBW No * NBW Joe Kwak, InterDigital Submission

PSNI Analysis: Relation to BER/FER PSNI is a direct measure of observed SNIR considering all channel impairments and implementation losses measured at the demodulator. PSNI is specified with respect to output FER, which considers all implementation losses including any FEC decoder implementation loss. Each STA will measure PSNI using a correction factor Cfy to account for the actual coding gain (CGact) of each FEC decoder. Any STA measuring PSNI on a frame using FEC will use CFy so that the reported PSNI from all STAs is normalised and assumes a theoretical coding gain. CFy = CGtheo - CGact = actual FEC decoder loss Reported PSNI value may be used to estimate OEbNo and BER/FER (QOS) for the reporting STA for any data rate. OEbNo = (PSNI/6) + 4.4dB - DRMx : For data rates without FEC decoder, OEbNo is used with the theoretical EbNo curve for that modulation to estimate BER. For data rates with FEC decoder, OEbNo is used with the theoretical FEC EbNo curve for for that modulation to estimate BER. Note: PSNI relation to BER is specified in AWGN and fading. In fading channels, the mean of the measured parameter used as basis for PSNI is adjusted using the measured variance to align the mean in fading with the AWGN case, as demonstrated for EVM in the simulations described in 802.11-03/773r2. Joe Kwak, InterDigital Submission

Why Does Variance Measurement Permit Alignment with AWGN? 1. Fading channels produce dynamic variations in instantaneous SNR 2. Range of individual packet SNRs increases with increasing variance. 3. For same aggregate effect on BER, the mean SNR of a fading channel must be higher than AWGN for low BERs. 4. For same aggregate effect on BER, the mean SNR of a fading channel must be lower than AWGN for high BERs. Joe Kwak, InterDigital Submission

Why Does Variance Measurement Permit Alignment with AWGN? 5. The variance of the measured parameter can be used to compute an adjustment factor to align the mean values in fading channels to the mean value in AWGN. 6. In this way the PSNI indicator relates directly to the BER/FER performance curves for AWGN in all channel conditions. Joe Kwak, InterDigital Submission

Variance changes slope of BER curve [Base chart from Brian Johnson, 802.11-03/682r0] Joe Kwak, InterDigital Submission

Variance changes slope of BER curve Joe Kwak, InterDigital Submission

Variance effects Data BER vs IEVM (with FEC) -1.00 -5.00 -1 0.00E -1.00 -5.00 -1 0.00E 5.00E 1.00E 1.50E 2.00E 2.50E E+01 E+00 +00 +00 +01 +01 +01 +01 -2 -3 BER(out of FEC decoder) (10-x) -4 -5 -6 -7 -8 BPSK, R=1/2, AWGN 20*log10(1/EVM) Joe Kwak, InterDigital Submission

EVM variance aligns fading BER with AWGN BER Joe Kwak, InterDigital Submission

Variance factor can align fading with AWGN Joe Kwak, InterDigital Submission

Variance factor can align fading with AWGN Joe Kwak, InterDigital Submission

Study of Sample Size for Error Analysis In fading channels, IEVM varies significantly from packet to packet. How many packets need to be measured to get a useful IEVMmean and IEVMsd? 50nsec fading channel was simulated for all OFDM rates using different sample sizes of 10, 100, 1000 packets per measurement Joe Kwak, InterDigital Submission

Sample Size Study Results Partial results for BPSK, R=1/2 Joe Kwak, InterDigital Submission

Sample Size Study Results (cont) Std Dev of IEVMmod measurements : max : min Joe Kwak, InterDigital Submission

Sampling error for IEVM Based on these results we see that normal sample error may have significant effect on IEVMmod. In AWGN-dominated channels, a single EVM measurement is adequate. In fading channels where EVM variance is high, a large number of measurements are required to achieve accurate results: When measured over 100 packets, IEVMmod results may vary over a +/- 2db range. When measured over 1000 packets, IEVMmod results may vary over +/- .4dB range. IEVMavg, IEVMsd and adequate sample size are all needed for meaningful measurement. Joe Kwak, InterDigital Submission

PSNI is Specified for “Black Box” SET INPUTS READ OUTPUT Sig Level Chan Fade PSNI STA RCVR (at selected rate) Noise BER or throughput RCVR implementation may use: -EVM on data symbols -EVM on pilots -other FEC decoder metrics -FEC corrected bit rate -spreading code correlation quality -many others possible Adjust sig, noise, fading for desired BER or Throughput for selected rate Joe Kwak, InterDigital Submission RCVR implementation may use: -EVM on data symbols -EVM on pilots -other FEC decoder metrics -FEC corrected bit rate -spreading code correlation quality -many others possible

PSNI Normative Specification Text The PSNI indicator is a measure of the perceived, post-processing signal-to-noise-plus-interference (S/(N+I)) ratio in the demodulator. The allowed values for the Perceived Signal to Noise Indicator (PSNI) parameter shall be an 8 bit value in the range from 0 through 255. This parameter shall be a measure by the PHY sublayer of the perceived signal link quality observed after RF downconversion and is derived from internal digital signal processing metrics of the demodulator used to receive frames on that link. PSNI shall be measured over the PLCP preamble and over the entire received frame. PSNI is intended to be used in a relative manner, and it shall be a monotonically increasing, logarithmic function of the observed link S/(N+I). Specified PSNI performance shall be measured over no less than 1000 PPDUs from the same transmitter. PSNI accuracy and range shall be specified in AWGN and fading for each data rate as follows: Theoretical FEC coding gain assumed in FER calculations: R = 1/2, 5.4dB gain R = 2/3, 4.7dB gain R = 3/4, 4.4dB gain PSNI SPECIAL VALUE: “0” shall indicate inability to measure PSNI When PSNI exceeds high end of measurable range for a given data rate, maximum PSNI for that rate shall be reported. Fading channel model is IEEE exponential ray decay with 50nsec decay time. Joe Kwak, InterDigital Submission

Conclusions RSSI is inadequate indicator of link quality. RCPI quantifies power levels at input to receiver. PSNI quantifies observed link quality at output of receiver. EVM or EVMavg, without knowledge of channel condition, cannot accurately indicate observed link quality in fading channels. Need variance adjustment, as demonstrated. PSNI premise has been validated. EVM on data symbols (with variance adjustment) is adequate basis for PSNI. But as Steve Pope, Carl Andren and others have indicated, other demod parameters may be preferred by certain manufacturers. PSNI is the only link quality proposal in TGk which can indicate output quality (BER/FER) for all rates and channel conditions. PSNI does not constrain manufacturers to particular implementation. Without a quantified link quality measurement, TGk’s work is not finished. PSNI is mature enough (10 months) for inclusion into TGk draft. Joe Kwak, InterDigital Submission

Motion for PSNI normative text Move to instruct the editor to incorporate the text from document 11-03-958r1-K-PSNI_NormText.doc into the TGk draft specification document Moved by Joe Kwak Seconded by: _______________ Vote YEA _______ Vote NEA _______ ABSTAIN _______ Motion Passes/Fails at ___% Joe Kwak, InterDigital Submission