Doc.: IEEE 802.22-08/0058r0



IEEE P802.22

Wireless RANs

|IEEE 802.22.1 Fades and Interference from IEEE 802.22 Networks |

|Date: 2008-02-01 |

|Author(s): |

|Name |Company |Address |Phone |email |

|Stephen Kuffner |Motorola |Schaumburg, IL |847-538-4158 |Stephen.Kuffner@ |

| | | | | |

Average Channel Gain over a Bandwidth

For a given mulitpath model, the ratio of the fade in the TG1 transmitter-to-WRAN sensing receiver path F↓, integrated over the TG1 77 kHz channel bandwidth, can be compared to the maximum upfade F ↑ (integrated over the microphone receiver bandwidth of 200 kHz) in the WRAN transmitter-to-microphone-receiver path. It should be noted that this comparison is dependent on the channel model, in particular the order of magnitude of the delays. Figure 1 shows a plot of a channel using WRAN model B [1] with the nominal delays (randomized by a perturbation between ± 100 ns to remove frequency domain periodicity). The blue curve shows a channel response for a given draw of the ray random phases for the nominal delays, while the red curve shows the channel response for a given draw of the ray random phases for 1/5th the nominal delays. The bright green lines show the regions of integration for the 77 kHz-wide TG1 channel (“beacon BW”) and one of the 200 kHz-wide regions for determining the maximum upfade. As can be seen, the longer (nominal) delay spreads result in much finer grain frequency variations and hence averaging over a given bandwidth will reduce the extremes due to smoothing, while for the shorter delays (nominal/5), there will be less smoothing, hence one can expect greater extremes. This is in fact shown in the statistical analysis in Figures 2 and 3.

[pic]

Figure 1. Example channel responses for the nominal WRAN channel B (blue) and channel B with delays scaled by 1/5th (red).

[pic]

Figure 2. Averaged channel gains vs. trial number. Note this is not channel gain vs. frequency. The blue line is the TG1 77 kHz-averaged channel gain, the red the maximum of all of a TV channel’s 200 kHz bandwidth averaged channel gains. Note the TG1 gain can exceed the maximum 200 kHz channel gain since it averages over a narrower bandwidth and so applies less smoothing to larger narrow peaks.

[pic]

Figure 3. Average channel gains vs. trial number with nominal channel B delays divided by 5, giving more extreme max values and fades.

The average channel gains above are calculated as [pic], where [pic] and [pic], where the hi are the power-normalized WRAN B ray magnitudes. The B is the averaging bandwidth; for the TG1 channel B = 77 kHz and for the interference channel B = 200 kHz. Figure 5 compares the TG1 77 kHz averaged fade depth to the cumulative density for a true flat Rayleigh fade ([pic], where F is the fade depth in dB). There is about 4 dB difference at the 5% points.

[pic]

Figure 4. Depth of TG1 fade for channel B profile averaged over the 77 kHz beacon bandwidth.

[pic]

Figure 5. Cumulative density for depth of averaged TG1 WRAN-B fade compared to true Rayleigh flat fade.

[pic]

Figure 6. Ratio of TG1 fade to max interference channel gain for nominal delays. Five percent of the cases have a ratio greater than 11.2 dB, while 1% have a ratio greater than 15.7 dB.

[pic]

Figure 7. Ratio of TG1 fade to max interference channel gain for nominal delays divided by 5. Five percent of the cases have a ratio greater than 15 dB.

As was shown in Figure 1, the shorter delay spread condition leads to more slowly varying (over frequency) channel responses and more extreme differences in average channel gain, by about 4 dB for the WRAN channel B nominal T vs. T/5 condition analyzed here.

Use of the Statistics

How these statistics relate to the protection of low power secondary licensed devices requires further consideration. It is not simply the probability of a fade of a certain depth that will really matter here. For example, if it is 1% PER that is required at the WRAN sensor, then many combinations of probability of a fade and probability of a packet error could result in 1%. For example, if the channel was a simple two-state channel, faded by 15 dB 10% of the time or 0 dB the remaining 90% of the time, a nominal SNR of 20 dB would give 5 dB SNR 10% of the time and 20 dB SNR 90% of the time. If the PER at 5 dB SNR was e.g. 10% and the PER at 20 dB SNR was e.g. 0.01%, then the expected value of the PER would be 10% x 10% + 90% x .01% = 1.009% so in this case a 10% PER in the faded condition is adequate for 1% average PER.

Instead of having the simple two-state channel as described in the preceding paragraph, the actual channel occupies a continuous density. What is required then is to find the nominal SNR value SNRnom that satisfies the following equation:

[pic] (1)

where X is the random variable for the SNR and p(X | SNRnom) is the probability density of the faded SNR given the nominal SNR. Note this simple analysis does not account for any ISI that may be present due to the faded channel frequency response. This simple analysis is only valid for flat fading. The required value for the average PER needs to be defined to enable determination of the nominal SNR.

For the simple two-state fading channel example above,

[pic] (2)

Equation (1) is a complicated integral to solve in general, but it need not be solved at all because this is essentially what the fading channel PER simulations do, and they further account for ISI that is not accounted for by a simple flat fading analysis. The PER simulations give the average PER for a given SNRnom over many random draws of the channel, some percentage of which are 10 dB fades, some percentage of which are 20 dB fades, etc. Thus, the WRAN channel B PER simulation results for MSF2 (the longest unprotected packet) show below in Figure 8 that 17.6 dB chip SNR (26.6 dB symbol SNR) is required for SNRnom to satisfy 1% average PER. Non-faded PER results for all packet types using the frame structure from the most recent draft standard are shown in Figure 9 for reference [3].

Conclusion

While the statistics for the averaged channel gain over the bandwidth of the beacon can be determined, these statistics do not necessarily reflect the performance of the channel since they imply flat fading that does not account for intersymbol interference.

[pic]

Figure 7. PER vs. chip SNR for AWGN and WRAN B for uncoded MSF2. There is a 12.8 dB difference between AWGN (4.8 dB) and WRAN B (17.6 dB) 1% PER chip SNR values. An Ec/No of 22.5 dB is required for the flat Rayleigh channel. Based on 100 packet errors.

[pic]

Figure 8. AWGN PER for various TG1 packets (from [2]) based on the most recent draft standard [3]. The 51-octet uncoded MSF2 is the worst case, shown here requiring about 5.5 dB for 1% PER; note there is about a 0.7 dB difference from the 4.8 dB of Figure 7.

References

[1] IEEE 802.22-05/0055r7, “WRAN Channel Modeling,” E. Sofer, G. Chouinard, 30 Aug 2005.

[2] IEEE 802.22-08/0zzzr0, “Simulation Results for TG1 According to Draft 2.0,” Wu Yu-chun, Jan 2008.

[3] P802.22.1/D2, “Part 22.1: Enhanced Protection for Low-Power, Licensed Devices Operating in Television Broadcast Bands,” October 2007.

-----------------------

Notice: This document has been prepared to assist IEEE 802.22. 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 grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE 802.22.

Patent Policy and Procedures: The contributor is familiar with the IEEE 802 Patent Policy and Procedures

, including the statement "IEEE standards may include the known use of patent(s), including patent applications, provided the IEEE receives assurance from the patent holder or applicant with respect to patents essential for compliance with both mandatory and optional portions of the standard." Early disclosure to the Working Group of patent information that might be relevant to the standard is essential to reduce the possibility for delays in the development process and increase the likelihood that the draft publication will be approved for publication. Please notify the Chair as early as possible, in written or electronic form, if patented technology (or technology under patent application) might be incorporated into a draft standard being developed within the IEEE 802.22 Working Group. If you have questions, contact the IEEE Patent Committee Administrator at .

Abstract

The statistics for the average fade over the 77 kHz TG1 bandwidth, compared to the maximum interference gain from a WRAN device into a microphone receiver, averaged over all 200 kHz microphone receiver subchannels within a TV channel, are determined using the WRAN channel B power/delay profile. For 95% of the cases, the beacon channel is faded less than 11.2 dB relative to the maximum interference channel, while for 99% of the cases the beacon channel is faded less than 15.7 dB relative to the maximum interference channel.Y[`bfghijlmnyzœ«¬´µÃÄÐÑîïðñòóô | ? ‚ [?]¶¸¼˜

¢

¨

É

Ë

8

F

J

The utility of these statistics is still to be determined, since they do not reflect degradations in beacon sensitivity due to ISI from frequency selectivity over the beacon bandwidth. Lastly, the average PER is identified as a parameter that needs to be defined and determined. One definition is proposed here.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download