Appendices 2-4. Utilisation of key licence exempt bands and the effects on WLAN performance. Issue 1 June Prepared by:

Transcription

1 Utilisation of key licence exempt bands and the effects on WLAN performance Appendices 2-4 Issue 1 June 2013 Prepared by: MASS Enterprise House, Great North Road Little Paxton, St Neots Cambridgeshire, PE19 6BN United Kingdom T: +44 (0) F: +44 (0) E: systems@mass.co.uk W: Page 1 of 46

2 Contents Appendix 2 Results of Fine Time resolution monitoring 3 1 Survey E02 - Shopping centre, Stratford, London Shopping centre, coarse time resolution Shopping centre, fine time resolution 7 2 Survey E03 Café, Shaftesbury Avenue, London Café, coarse time resolution Café, fine time resolution 11 3 Survey E04 Apartment, Wandsworth Apartment, coarse time resolution Apartment, fine time resolution 15 4 Survey E05 house, Millwall House, coarse time resolution House, fine time resolution 19 Appendix 3 Monitoring System Specifications 21 1 Monitoring systems 21 2 Recommendations for future system specifications 23 Appendix 4 Stress Test Results 25 Page 2 of 46

3 APPENDIX 2 RESULTS OF FINE TIME RESOLUTION MONITORING This appendix presents the results of monitoring with finer time resolutions at four sites. These sites were selected from the coarse time resolution monitoring survey sites listed in Appendix 1. The rightmost column in the table below gives the corresponding coarse time resolution survey to assist with comparison with the results in Appendix 1. Site type Survey number Location Appendix 1 Shopping centre E02 Stratford, London S01 Café E03 Shaftesbury Avenue, London S10 Apartments E04 Wandsworth S14 House E05 MiIlwall - For each survey the coarse resolution plots (summary time plot, summary frequency plot and spectrogam) are provided as in Appendix 1. For the fine time resolution results the following time plots are also provided for the primary and secondary parameters: Entire 2.4 GHz band over all the time recorded; Selected channel in 2.4 GHz band over all time recorded; Selected channel in 2.4 GHz band over the busiest hour. Page 3 of 46

4 Entire 2.4 GHz band over all the time recorded These four plots show the results from two colocated monitoring systems. The first system was configured for hourly averaging and was identical to the system used for collecting the results in Appendix 1. This system was running in the fixed location mode, as described in Appendix 3. In the second system the same scan strategy was used but the data was not averaged across scans, so each data point on the graph represents 330 seconds in time. This system was running in the mobile mode, as described in Appendix 3. Occupancy in the physical layer across the entire 2.4 GHz band averaged in hourly (red) and 10 second intervals (blue) MAC stress estimated from the percentage retries and averaged in hourly (red) and 10 second intervals (blue) The busiest hour in the 2.4 GHz band has been taken from the time of hourly averaged peak throughput Throughput in the MAC layer in hourly (red) and 10 second intervals (blue) Network density measured by counting the number of unique BSSIDs in one hour (red) and in 10 seconds (blue) Selected channel in 2.4 GHz band over all time recorded These plots are laid out in the same way as the previous plots. They show the same four parameters with averaging in hourly and 330 second intervals. The difference here is that a single channel has been selected for analysis. In each case the selected channel is stated in the text and has been selected as it was expected to carry the most traffic. Selected channel in 2.4 GHz band over the busiest hour The plots are arranged similarly to the previous two sets of plots. Here the green trace has been added. This additional trace shows the results from a third monitoring system that was continuously recording on the selected channel during the time that it was expected to be most busy at that location. This system was running in the single channel mode, as described in Appendix 3. The averaging time for the third trace was 5 seconds. Page 4 of 46

5 1 SURVEY E02 - SHOPPING CENTRE, STRATFORD, LONDON 26 February 2013 The shopping centre location for this survey corresponds to S01 in Appendix GHz 5 GHz Occupancy Moderate Low MAC stress Moderate Moderate Throughput Very low Very low Network density High High Diurnal patterns The overall pattern of activity was very similar to the last time this site was surveyed (26 October 2012) with the peaks in occupancy and throughput occurring in the afternoon and early evening. The rise in the 5 GHz network density seen previously was not seen in this case. Other observations The MAC stress is moderate in the 5 GHz band and there are peaks in the lower of the two 5GHz sub-bands that suggest intermittent use of a technology other than WiFi. This effect is also seen in the café and house surveys, suggesting a common factor in this second survey phase. The high network density was the most noticeable facet of this survey. The different algorithms and monitoring systems used did give markedly different estimates of the number of APs, but, irrespective of the algorithm used for estimation, the network density would be classified as high. The more interesting point about the network density to note was the fact that the number of APs varied throughout the survey. The most likely explanation for this was very large numbers of tethered mobile phones and/or dedicated mobile hotspots devices. There are few studies of the prevalence of tethered phones/ mobile hotspots, but there is some anecdotal evidence to support it (Chernicoff, 2011). The use of a finer time resolution revealed the variations in each of the parameters within the busiest hour. If the highest peak in the 5s averaged data had been used for classification then the 2.4 GHz band occupancy and MAC stress would have increased from moderate to high. The throughput would have remained very low and the network density would still be high. Page 5 of 46

6 1.1 Shopping centre, coarse time resolution Page 6 of 46

7 1.2 Shopping centre, fine time resolution Entire 2.4 GHz band over all time recorded Page 7 of 46

8 Channel 11 over all time recorded Channel 11 over busiest hour Page 8 of 46

9 2 SURVEY E03 CAFÉ, SHAFTESBURY AVENUE, LONDON 4 March 2013 The location for this survey was previously visited on 12 November See Appendix 1 survey S10 for the results of that survey. 2.4 GHz 5 GHz Occupancy Moderate Low MAC stress High Moderate Throughput Low Very low Network density Low Low Diurnal patterns Other observations In this survey the peak activity was in the middle of the day unlike the previous survey where the peak was in the evening. The radio environment in this survey was very similar to that seen in the previous survey with moderate 2.4 GHz occupancy and high MAC stress plus moderate MAC stress in the 5 GHz band. The throughput in both bands increased a little. Again the high level of MAC stress is attributed to the use of overlapping channels at the lower end of the 2.4 GHz band. If the peak of the fine time resolution data had been used for classification rather than the hourly data, none of the classifications would have changed. 2.1 Café, coarse time resolution Page 9 of 46

10 Page 10 of 46

11 2.2 Café, fine time resolution Entire 2.4 GHz band over all time recorded Channel 1 over all time recorded Page 11 of 46

12 Channel 1 over busiest hour Page 12 of 46

13 3 SURVEY E04 APARTMENT, WANDSWORTH 17 March 2013 The location for this survey was previously surveyed as S14 in Appendix GHz 5 GHz Occupancy Low Low MAC stress Low Low Throughput Low Very low Network density Low Low Diurnal patterns Other observations Evening peaks in throughput. This survey was very similar to the previous survey at this site. There was a little more activity in the 5 GHz band but, otherwise, the radio environment was unchanged. Examining the finer time resolution in the busiest hour does not reveal dramatic differences from the hourly averaged data. The peak occupancy would be moderate rather than low. MAC stress was generally low but was occasionally moderate and was high for a single 5 s period. The throughput and network density were both low throughout regardless of the measurement interval. 3.1 Apartment, coarse time resolution Page 13 of 46

14 Page 14 of 46

15 3.2 Apartment, fine time resolution Entire 2.4 GHz band over all time recorded Channel 11 over all time recorded Page 15 of 46

16 Channel 11 in busiest hour Page 16 of 46

17 4 SURVEY E05 HOUSE, MILLWALL 28 March 2013 This site was not surveyed during the first phase of work so there are no corresponding results in Appendix 1. There was some data corruption which meant that the 5 s averaged data in the busiest hour was lost. 2.4 GHz 5 GHz Occupancy Moderate Low MAC stress Low n/a Throughput Low Very low Network density Low None Diurnal patterns Other observations Although a number of APs were visible at this site there was very low utilisation and no clear patterns of usage. Both the occupancy and the throughput in the 2.4 GHz band were bursty at this location. The MAC stress increased to moderate in the busiest hour but only in one 10 s sample, so it is rated as low from the hourly averaged results. Broadband interference was present for 10% of the time but it was below -80dBm and does not seem to have affected the WiFi traffic. Had there been more utilisation of the WiFi networks at this site then it is possible that the broadband interference could have had some impact on the MAC stress. The APs are not using overlapping channels so the neighbours here have adopted the recommended way of configurating their networks to minimise interference between WiFi channels. Page 17 of 46

18 4.1 House, coarse time resolution Page 18 of 46

19 4.2 House, fine time resolution Entire 2.4 GHz band over all time recorded Page 19 of 46

20 Channel 1 over all time recorded Page 20 of 46

21 APPENDIX 3 MONITORING SYSTEM SPECIFICATIONS This appendix details the specifications of the monitoring systems used for coarse time resolution monitoring, fine time resolution monitoring and our recommendations for a future monitoring system. 1 MONITORING SYSTEMS Each monitoring system comprised a Windows 7 laptop computer fitted with a GPS receiver and two dongles (Figure 1) which were: AirPcap Nx WiSpy DBx Figure 1 Laptop fitted with dongles and GPS receiver Software was written to run on the laptop which communicated with the GPS receiver and dongles and automatically saved data to disk in a CSV file format. Page 21 of 46

22 Figure 2 User interface of the capture software The monitoring system was run in one of three modes which were designed to gather information to support the experiments needed for this study. In coarse time resolution monitoring only the fixed location mode was used. All three modes were used for fine time resolution monitoring. The main parameters of these modes were as follows: Fixed location mode Mobile mode Single channel mode Time resolution Coarse Fine Fine Averaging interval 1 hour 330 s 5 s Channel dwell time 10 s 10 s 5 s AirPcap scan time 330 s 330 s Not scanning Bands scanned 2.4 & 5 GHz 2.4 & 5 GHz 2.4 GHz MAC frame recording (note 1) AirPcap 20/40 mode (note 2) No No Yes 20MHz 20MHz 20MHz Note 1 Recording all MAC frames yields the best fidelity information but consumes considerable disk space. For this project MAC frames were only recorded when the monitoring system was set into single channel mode for the purposes of investigating the busiest hour. Note also that user data was removed from the MAC frames by the monitoring software prior to recording. Page 22 of 46

23 Note 2 The AirPcap dongle cannot simultaneously report 20MHz and 40MHz WiFi traffic, it has to be set into the required mode. For this study the device was left in its 20MHz mode throughout and we relied on the WiSpy results to observe the presence of 40MHz traffic. Future systems should be designed to improve this situation and monitor wider band channels comprehensively at the MAC layer. In all modes the WiSpy scan settings were as follows. 2.4 GHz 5 GHz IR2006A 5GHz IR2006B Start frequency 2400 MHz 5160 MHz 5480 MHz Steps Step size khz khz khz Dwell time per step 1 ms 0.25 ms 0.25 ms Filter bandwidth khz 375 khz 375 khz 2 RECOMMENDATIONS FOR FUTURE SYSTEM SPECIFICATIONS Based on the experience of the prototyping exercise carried out in this study, future monitoring systems should have the following characteristics: Characteristic Bands Antennas Modes Recording duration Busiest hour recording Channel dwell time, MAC layer Averaging interval Monitored parameters Recommendation 2.4 GHz & 5 GHz Internal and external omnidirectional, selectable by field staff Static location and roaming Static location mode At least one month to allow diurnal and weekly variations to be observed Roaming mode At least two days to allow diurnal variations to be observed, especially determination of busiest hour In both modes the system should record all monitored parameters during the busiest hour with a 5 second averaging interval. 5 seconds 15 minutes for all parameters across entire recording duration 5 seconds during busiest hour Occupancy (PHY) MAC stress (MAC) Throughput (MAC) Network density (MAC) Traffic type (MAC) The current system saves data into CSV files as these were simple to produce and analyse for the purposes of this research study. We recommend that future work includes standardisation of file formats that can be exchanged readily with other researchers. Page 23 of 46

24 The experiments carried out during this study have revealed some of the issues surrounding the use of low cost hardware for spectrum monitoring in the LE bands. Of particular note are the following issues: 1. There is no common agreement on the use of internal or external antennas when monitoring with dongles. We have taken the stance for this study that the internal antennas can be taken as being representative of typical applications in laptops and mobile phones. However sometimes the field team had to set the monitoring systems up inside cupboards for security purposes and, in those situations, there would be a very strong argument for using external antennas; 2. Much of the readily available, low-cost hardware and software for recording, processing and analysing traffic in the LE bands assumes that the user data can be intercepted and recorded. This is not permissible in the UK unless explicit permission has been obtained from all the users, which is wholly impractical in the urban environment. As with the system used here, future monitoring systems should therefore be designed to remove the user data from all intercepted traffic at the point of detection; 3. Additional processing may be needed to refine the data coming from such dongles. In this study considerable effort went into removing ghost access points from the raw AirPcap data. The team found that this device reports APs that appear in channels other than the one to which an AP is actually transmitting. The post-processing code runs an algorithm that removes all APs that appear on the channels other than the primary one. The decision as to which is the primary channel is made on the observed bit rate. Page 24 of 46

25 APPENDIX 4 STRESS TEST RESULTS This appendix details the results of carrying out stress testing of the 2.4 GHz band at Ofcom s headquarters in Riverside House, London. Testing involved streaming data on three WiFi networks, as shown in the figure below. In addition to streaming data over the WiFi networks an analogue video sender was used as a source of potential interference. Monitoring was performed using the system developed in the first part of the project. The monitoring software was running in its fixed location mode in which one hour averaging was used. Figure 3 Stress test equipment configuration (figure courtesy of Ofcom) Page 25 of 46

26 The following table gives the test configurations used. Test Category WiFi Channels FTP Data Rate Test reference (note 1) Non-overlapped channels (note 2) 1, 5, Mbps A (F02) 20 Mbps B (F01) 1, 5, 13 + Analogue 20 Mbps C (F03) Co-channel 5 (3 off) 1.2 Mbps D (F10) 20 Mbps E (F09) Two adjacent channels 4, 5, 6 One adjacent channel 4, 5, Mbps F (F05) and G (F06) 20 Mbps H (F04) 1.2 Mbps I (F08) 20 Mbps J (F07) Note 1: Tests have been relabelled to help clarify the test results. The reader should use the alphabetical labels (A, B, C...). The labels in parentheses (F01, F02,...) have been retained for the study team s use. Note 2: In the tests specified by Ofcom the WiFi channels 1, 5, 9 and 13 were identified as nonoverlapping. Please refer to section 2 of this report where the issue of channel allocation in the 2.4 GHz band is discussed and where it is recommended that 1, 6 and 11 are used to minimise the probability of interference between networks. The detailed results are given in the following pages. The main points are summarised here to aid with assimilation. Test Category WiFi Channels FTP Data Rate Reference Occupancy MAC stress Throughput Non-overlapped channels 1, 5, Mbps 20 Mbps A Moderate Low Low to moderate B High Low Moderate to high 1, 5, 13 + Analogue 20 Mbps C High Low Moderate to high Co-channel 5 (3 off) 1.2 Mbps D Moderate Moderate Low to moderate 20 Mbps E High Moderate Moderate Two adjacent channels 4, 5, Mbps F Moderate Moderate Low to moderate G Moderate Moderate Low to moderate 20 Mbps H High Moderate Moderate to high One adjacent channel 4, 5, Mbps I Moderate Moderate Moderate 20 Mbps J High Low Moderate to high Page 26 of 46

27 Test A (F02) Low speed FTP traffic, 2.4 GHz channels 1, 5, GHz 5 GHz Occupancy Moderate Low MAC stress Low Low Throughput Low to moderate Very low Network density Low Low Observations The occupancy in the 2.4 GHz band was only about 15%, so considerably lower than tests B and C. The MAC stress in this band was low indicating that this configuration of channels supports low to moderate throughputs without problems. Page 27 of 46

28 Page 28 of 46

29 Test B (F01) High speed FTP traffic, 2.4 GHz channels 1, 5, GHz 5 GHz Occupancy High Moderate MAC stress Low Low Throughput Moderate to high Very low Network density Low Low Observations Compared to test A, here the FTP data rate was increased from 1.2 Mbps to 20 Mbps. This led to an increase in occupancy which exceeded 30% and was therefore rated as high. Channel 13 was in use for more than 50% of the time and channel 1 was nearly as busy. Channel 5 overlaps both of these a little but was only active for about 10% of the time. It is reasonable to conclude that the interaction between channel 5 and the other two channels caused the MAC stress to rise to a moderate level. Page 29 of 46

30 Page 30 of 46

31 Test C (F03) High speed FTP traffic, 2.4 GHz channels 1, 5, 13 + Analogue 2.4 GHz 5 GHz Occupancy High Moderate MAC stress Low Low Throughput Moderate to high Very low Network density Low Low Observations This test resembles test B closely until the point where the analogue video sender was turned on. At that point all WiFi traffic in the test networks ceased, which can be seen as a marked drop in the 2.4 GHz band throughput in the last four hours of the test. The presence of the analogue video can be seen clearly in summary plot below the channel 13 AP. The MAC stress did not increase due the presence of the analogue video sender, because the traffic ceased when it was active. This is one case where the MAC stress is not an effective indicator of MAC stress. Page 31 of 46

32 Page 32 of 46

33 Test D - (F10) Low speed FTP traffic, 2.4 GHz channel 5 x GHz 5 GHz Occupancy Moderate Low MAC stress Moderate Low Throughput Low to moderate Very low Network density Low Low Observations This was a similar configuration to test A, but in this case all the activity was on the same channel. Channel 5 was used less than 50% of the time and the overall occupancy of the band was rated as moderate. MAC stress increased from low in test A to moderate here. This indicates that the MAC is starting to show signs of having to share the channel 5 resource between the networks but is coping and the users should not notice any problems. Page 33 of 46

34 Page 34 of 46

35 Test E (F09) High speed FTP traffic, 2.4 GHz channel 5 x GHz 5 GHz Occupancy High Low MAC stress Moderate Low Throughput Moderate Very low Network density Low Low Observations Channel 5 was in use over 50% of the time which is the main reason for the occupancy being high. Despite the multiple usage of a single channel the MAC stress only just came into the moderate category and was below 8% throughout. Comparing these results with those from test D the conclusion is that PHY is starting to show signs of heavier usage but it has not caused any significant effects in the MAC. Comparing tests D and E with A, B and C it will be seen that the overall throughput appears to have dropped and the MAC stress increased. Page 35 of 46

36 Page 36 of 46

37 Test F - (F05) Low speed FTP traffic, 2.4 GHz channels 4, 5, GHz 5 GHz Occupancy Moderate Low MAC stress Moderate Low Throughput Low to moderate Very low Network density Low Low Observations This test is very similar to test D, the difference being that three adjacent, overlapping channels have been used. The general understanding in the literature, as detailed in the main body of this report, is that this configuration should be less efficient than cochannel operation, as traffic in neighbouring channels will be regarded as noise by the CSMA/CA protocol in WiFi devices. Comparing the results with test D we observe that this test showed very little difference in occupancy and throughput. The MAC stress is also very similar, if anything it has dropped a little compared to test D. Such a result would be counter to the common understanding and likely to raise questions. Further repetitions of this test are therefore required to verify the result. Test G below is one such repetition. Page 37 of 46

38 Page 38 of 46

39 Test G - (F06) Low speed FTP traffic, 2.4 GHz channels 4, 5, GHz 5 GHz Occupancy Moderate Low MAC stress Moderate Low Throughput Low to moderate Very low Network density Low Low Observations The results of this test accord with test F. Together they would suggest that working on adjacent channels is better than co-channel operation. On their own these tests are not considered sufficient to overthrow the common understanding, but do raise the question of whether the channel allocation question should be revisited. Our opinion on this is that the results of tests E and H should take precedence over D, F and G as they are carrying higher throughputs. Page 39 of 46

40 Page 40 of 46

41 Test H - (F04) High speed FTP traffic, 2.4 GHz channels 4, 5, GHz 5 GHz Occupancy High Moderate MAC stress Moderate Low Throughput Moderate to high Very low Network density Low Low Observations This test should be compared with test E with the difference being the use of neighbouring channels rather than co-channel operation. The occupancy and throughput results are very similar. Looking at the MAC stress it is a little higher in this test than in test E. Unlike tests F and G, this result suggests that neighbouring channel operation is slightly worse than co-channel operation. As tests E and H had higher throughputs than D, F and G, the overall conclusion is that these tests provide some evidence towards using co-channel rather than adjacent channel operation. Page 41 of 46

42 Page 42 of 46

43 Test I (F08) Low speed FTP traffic, 2.4 GHz channels 4, 5, GHz 5 GHz Occupancy Moderate Low MAC stress Moderate Low Throughput Moderate Very low Network density Low Low Observations This test should be compared with test A, the difference being that channel 1 has been moved to channel 4 so that it overlaps channel 5. In both tests the occupancy was moderate and the throughput was low to moderate. In test I the MAC stress increased from below 2.5% (low) in test A to above 9% (moderate) in the busiest hour. This strongly suggests that this overlapping configuration of test I is worse than the non-overlapping configuration of test A. Page 43 of 46

44 Page 44 of 46

45 Test J (F07) High speed FTP traffic, 2.4 GHz channels 4, 5, GHz 5 GHz Occupancy High Low MAC stress Low Low Throughput Moderate to high Very low Network density Low Low Observations Despite a high occupancy and moderate to high throughput in this test the MAC stress was low at just over 3% in the busiest hour. Comparing this result with test A and test I suggests that a loaded 2.4 GHz band in this configuration behaves in a similar way to the non-overlapping scenario of test A and is better than the unloaded scenario of test I. This result appears anomalous. The peak in throughput occurred at the very end of the test and it is possible that the monitoring system did not see a complete hour of maximum throughput. There was insufficient time on the project to repeat the test. Page 45 of 46

46 Page 46 of 46