Predictable 802.11 packet delivery from wireless channel measurements
Summary (4 min read)
1. INTRODUCTION
- Wireless LANs based on 802.11 are used almost everywhere, from airports to zoos and in urban, suburban and rural areas.
- For good performance, reliability and coverage, the physical layer settings should match the RF channel over which the wireless signals are sent.
- Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.
- In practice, 802.11 LANs have never used channel measurements as more than a coarse indicator of expected performance.
- This approach is very effective for slowly varying channels and simple configurations (e.g., a few rates with fixed transmit power and channel) since the best setting will soon be found.
2. MOTIVATION
- Existing predictions of packet delivery for a given link are based on its Received Signal Strength Indication (RSSI) value.
- Convolutional coding is applied across the bits for error correction and bits are interleaved to spread them in frequency.
- Note that the details of single-stream 802.11n differ slightly from 802.11a/g (optimized coding rates and more data subcarriers), but in ways that are not material for their work so that the authors can treat 802.11n as a superset of 802.11a/g. Packet Delivery versus RSSI/SNR.
- This is consistent with other reported 1We refer to the metric computed from RSSI and noise measurements as the packet SNR, RSSI-based SNR, or simply RSSI.the authors.
- Many possible factors cause the observed variability for real channels, including NIC calibration, interference, sampling, and multipath.
3. PACKET DELIVERY MODEL
- The authors goal is to develop a model that can accurately predict the packet delivery probability of commodity 802.11 NICs for a given physical layer configuration operating over a given channel.
- In practice, the interesting regions for the four effective SNRs do not overlap because at a particular effective SNR value only one modulation will be near the transition from useless (BER ≈0.5) to lossless (BER ≈0).
- When graphs in this paper are presented with an effective SNR axis, the authors use all four values, each in the appropriate SNR range.
- Different devices may have different noise figures, a measure of how much signal strength is lost in the internal RF circuitry of the NIC.
- The transmitter also needs up-to-date CSI: either from feedback or estimated from the reverse path.
4. TESTBEDS
- The authors conduct experiments using two stationary wireless testbeds deployed in indoor office environments, T1 and T2 (Figure 4).
- Each testbed covers a single floor of a multi-story building and has a variety of links in terms of maximum supported rate and line-of-sight versus multi-path fading.
- The authors conduct mobile experiments using laptops that interact with testbed nodes and are configured in the same way.
4.1 Node Configuration
- Each node is a stationary desktop or portable laptop equipped with an Intel Wi-Fi Link 5300 (iwl5300) a/b/g/n wireless network adapter.
- They run the Linux 2.6.34 kernel with a modified version of the iwlagn driver [2].
- The authors use up to three transmit and receive antennas, supporting up to three MIMO streams, and the rates in Table 1 per stream.
- Each testbed operates on a 5 GHz channel unoccupied in its environment; there was no noticeable interference.
4.2 Measurement Tools
- The authors hardware enables us to vary the transmit power level from −10 dBm to 16 dBm in steps of 0.5 dB, and divides power equally across streams.
- These combine to define the per-receive-chain packet SNR : = RSSI (dBm)− Noise (dBm)− AGC (dB) (3) The iwl5300 calculates the quantities RSSI and Noise as the respective sums of average signal strength and average error vector magnitude in each OFDM subcarrier [2].
- This is exactly the traditional definition of SNR applied to OFDM.
- The authors configure the NIC to compute this feedback packet for every received frame, rather than just during sounding, and send it up to the driver instead of back to the transmitter.
- Each channel matrix entry is a complex number, with signed 8-bit resolution each for the real and imaginary parts.
5. PACKET DELIVERY EVALUATION
- The authors use their testbeds to experimentally evaluate how well their model of §3 predicts packet delivery.
- This is the fundamental measure of whether the model is useful; good predictions enable applications such as rate adaptation, transmit power control, antenna selection, and channel selection.
5.1 Measurement setup
- The authors first measure packet delivery for different antenna configurations over a 20 MHz channel on their testbeds.
- In each test, the authors send 1500 byte packets as constant bit-rate UDP traffic generated by iperf at 2 Mbps for 5 seconds.
- Note that CSI is measured during the preamble, so it does not depend on the transmit rate.
- Similarly, 3x3 CSI gives us the channel between each pair of transmit and receive antennas, so it also implicitly contains 1x1 CSI.
- The above testing gives us ground truth data to probe variation across 200 links, 26 dB of transmit power, four antenna configurations ranging from 1x1 to 3x3, and 8 per stream rates (for 24 rates with up to three streams).
5.2 RSSIs and Multiple Antennas
- The authors model predicts packet delivery in terms of effective SNR as described in §3.
- This is simple enough for the 1x1 case of a single transmit and receive antenna: the authors convert the single RSSI value to a packet SNR using Eq. (3), which is then mapped to packet delivery for the transmit rate that is used.
- The authors first convert the per-antenna RSSIs to SNRs and then sum the SNRs.
- This is a straightforward choice for a single spatial stream as it corresponds to receiver processing using MRC [8].
5.3 Results
- To compare their model with RSSI, the authors first analyze their 1x1 measurements to find the transition windows for all of the links in testbed T1.
- The authors define this to be the effective SNR or packet SNR values over which packet delivery rises from 10% to 90% for any link.
- While the transitions for the last four rates are inflated with RSSI, they remain tight with effective SNR.
- The larger significance of narrow transition windows is that, by reducing them enough that they do not overlap, the authors are able to unambiguously predict the highest rate that will work for all links nearly all of the time.
- They agree with the measured SNRs on a wired link (Figure 1(a)), which strongly suggests that the effective SNR captures the fundamental error characteristics of the link.
6. APPLICATION TO RATE SELECTION
- The most direct uses of packet delivery predictions are rate adaption, transmit power control, and channel selection.
- They provide a well-established baseline against which the authors can gauge their performance.
- The authors goal is to perform as well as the best, already nearoptimal 802.11a/g schemes on their home ground, with a method that has the advantages of simplicity, deployability, and generality.
- Rate adaptation is an open problem for 802.11n.
- Most schemes in the literature were not designed for MIMO systems, and none of the ones that were have been tested on real 802.11 channels.
6.1 Rate Selection Algorithms
- The authors experiment with ESNR, an algorithm based on their model, plus SampleRate [5], the de facto rate selection algorithm in use today, and SoftRate [28], a research algorithm with the best published results.
- It maintains delivery statistics for different rates to compute the expected airtime to send a packet, including retries.
- The input to these predictions is the bit error rate (BER) as estimated from side-information provided by the convolutional decoder.
- ESNR uses their model in a very simple way: given recent channel state information, compute the highest rate configuration that is predicted to successfully deliver packets (PRR > 90%).
6.2 Trace-driven Simulator
- 7 the authors turn to simulations to compare these algorithms.
- No algorithm will beat SampleRate by a significant margin on static channels, because it will quickly adapt to the channel.
- The CSI reflects frequency-selective fading over real, varying 20 MHz MIMO channels that is typically not observed with more narrowband experimentation, e.g., on the USRP.
- To ensure that ESNR is not given the unrealistic advantage of ground truth CSI, the authors corrupt the CSI at the level of ADC quantization, which typically induces an error of ±1.5 dB in the output effective SNRs.
- SoftRate estimates the BER directly during decoding.
6.3 Rate Adaptation Results
- The authors first examine the performance of ESNR for SISO rates.
- Even in these mobile channels, ESNR holds up very well and tracks Previous-OPT within 10%.
- Note that packet SNR was observed to fare quite poorly [28] in mobile channels, but since effective SNR reflects actual link quality its estimates are more accurate (§5) and stable (2–3× less variance).
- Next, the authors compare ESNR with SampleRate and SoftRate in Figure 11 and Figure 12.
8. CONCLUSION
- Wireless links are easy to understand in theory, but difficult to operate in practice, thus search is used to find the best rates, power levels, or other parameter of interest.
- The authors model takes as input the RF channel (measured as 802.11 Channel State Information) and predicts whether the link will deliver packets for a wide range of NIC configurations.
- It uses the notion of effective SNR to handle OFDM over faded links, works for MIMO configurations, and needs no calibration of target links.
- The authors show that, for the first time, measurements taken by commodity NICs can accurately predict whether links will work over a wide range of rates, transmit power, spatial streams, and antennas settings that have not previously been tested.
- In contrast, predictions based on RSSI often confuse from two to five rates as the potential best rate to use.
Did you find this useful? Give us your feedback
Citations
1,354 citations
Cites methods from "Predictable 802.11 packet delivery ..."
...It works on up-to-date Linux operating systems: in our testbed we use Ubuntu 10.04 LTS with the 2.6.36 kernel....
[...]
761 citations
704 citations
612 citations
Cites background or methods from "Predictable 802.11 packet delivery ..."
...Since current off-the-shelf CSIs are only available with Intel 5300 NIC and the modified driver [Halperin et al. 2010], it poses a challenge to employ CSIs on mobile handhelds....
[...]
...Leveraging the off-the-shelf Intel 5300 NIC and a modified driver, a group of sampled versions of CFRs within the WiFi bandwidth are revealed to upper layers in the format of Channel State Information (CSI) [Halperin et al. 2010]....
[...]
...In 802.11 a/g/n standards, channel response can be partially extracted from off-the-shelf OFDM receivers in the format of Channel State Information (CSI), which reveals a set of channel measurements depicting the amplitudes and phases of every subcarrier [Halperin et al. 2010]....
[...]
...DOI: http://dx.doi.org/10.1145/2543581.2543592...
[...]
...In fact, this sample version of CFR has been employed in recent adaptive wireless communication systems to improve reliability [Halperin et al. 2010] and throughput [Bhartia et al. 2011], as well as for precise indoor localization on off-the-shelf platforms [Wu et al. 2012; Sen et al. 2012a; Sen et…...
[...]
454 citations
References
8,084 citations
7,826 citations
"Predictable 802.11 packet delivery ..." refers background or methods in this paper
...This is a straightforward choice for a single spatial stream as it corresponds to receiver processing using MRC [8]....
[...]
...Textbook analyses of modulation schemes give delivery probability for a single signal in terms of the signal-to-noise (SNR) ratio [8], typically expressed on a log scale in decibels....
[...]
...This is a straightforward choice for a single spatial stream as it corresponds to receiver processing using MRC [8]....
[...]
...It is computationally simple, optimal and equivalent to Maximal-Ratio Combining (MRC) for a single stream, and near optimal for multiple streams....
[...]
...In IEEE PIMRC, 2002....
[...]
1,330 citations
Related Papers (5)
Frequently Asked Questions (17)
Q2. How can the authors evaluate the effectiveness of SNR?
The authors have also shown that effective SNR can be implemented on commodity NICs and evaluated it over real wireless channels with mobile and fixed clients.
Q3. What is the way to achieve good performance in a wide range of environments?
Modern wireless NICs provide a large and growing range of physical layer configurations to obtain good performance across this range of environments.
Q4. What is the way to handle OFDM over faded links?
It uses the notion of effective SNR to handle OFDM over faded links, works for MIMO configurations, and needs no calibration of target links.
Q5. What is the way to adapt to varying channel conditions?
Many rate adaptation algorithms have been proposed that use packet delivery statistics [5, 29], RSSI-based packetSNR [6, 10], or symbol-level details of packet reception [23, 28] to adapt to varying channel conditions.
Q6. What is the concept of an effective SNR for a multi-carrier channel?
Their model uses the concept of an effective SNR for a multi-carrier channel [18], such as OFDM, in which there are different subcarrier SNRs, plus approximations for coding, interference between MIMO streams, and decoding algorithms.
Q7. What is the method for calculating the supported rate on a link?
At each reduced transmit power level, the authors estimate the best supported rate on a link based on appropriate thresholds, and continue the reduction if the original rate is sustained.
Q8. What is the biggest potential weakness of the SNR protocol?
This is one of the largest potential weaknesses of this technique, because effective SNR is based on measurements taken only during the packet preamble.
Q9. How do the authors compute the effective SNR?
If the authors ignore coding for the moment, then the authors can compute the effective SNR by averaging the subcarrier BERs and then finding the corresponding SNR.
Q10. How do the authors measure the reception rate of a packet?
Then the authors collect packet reception rate (PRR) statistics for all 8 rates using 1, 2, and 3 spatial streams as the authors vary the power between −10 dBm and 16 dBm in steps of 2 dB.
Q11. What is the rate for a given SNR?
Except for extremely low and high SNRs, nearly all SNRs have at least two and up to five different rates as suitable choices for the best rate.
Q12. What does the study conclude that the presence of interference does not cause wild swings in transmit rate?
The authors conclude that the mere presence of interference does not completely invalidate effective SNR values, and thus transient interference will not cause wild swings in transmit rate.
Q13. What is the way to convert CSI to effective SNR?
The authors convert CSI to effective SNR in a way that better matches the equal modulation and power allocation used by 802.11n and offer a better API for practical use.
Q14. What is the way to understand how transition windows map to packet delivery predictions?
To understand how transition windows map to packet delivery predictions, the authors analyze their measurements for the highest supported rate (PRR≥ 90%) for each link and all NIC settings.
Q15. How can the authors predict the transmission power of a wireless link?
Their example in §5 suggests that, with a good predictive model, the authors can directly and confidently select a reduced transmit power without degrading link performance.
Q16. How is the Eff SNR line used to show that the link is tight?
To show that this trimming is tight, the authors also consider trimming towards slightly lower thresholds (Effective SNR− 0.5 dB, solid line).
Q17. How did the authors change the transmit power of the node designated as the interferer?
The authors also varied the transmit power of the node designated as the interferer from low to high to induce a large range of interfering channels.