Review: “WiFi-Nano: Reclaiming WiFi Efficiency Through 800 ns Slots”

Bhavesh Singh

2010CS50281

1. Summary

1.1 Motivation

As the speed of WiFi physical layer increases from 1 Mbps to 1 Gbps, the efficiency

reduces from over 80% to 10%. It would be revolutionary if such a decrease in efficiency could

be ceased or kept as minimum as possible. In 802.11, the reasons for such efficiency drop is

known. So WiFi-Nano has tried to overcome these reasons and to solve the problem behind

each reason. There are three key overheads which are responsible for this – channel access,

data preamble and acknowledgement overheads. Some other overheads like collision are

also there. At 600 Mbps, the average channel access overhead is 500% of the packet (1500

bytes long) transmission time. While at 1 Mbps, it is 0.83%. Similarly the differences in other

overheads are also evident. For example the data preamble and ACK overheads together

increase from 15% at 54 Mbps to 44% at 600 Mbps. Such differences cannot be ignored and

solutions to these problems will make the better usability of WiFi with more and more

physical layer data rates. Since the issue of WiFi efficiency is bigger, it makes these problems

very significant which WiFi-Nano tried to solve.

600 Mbps Overhead ~91%

1.2 Contribution

The main contribution was to improve the efficiency of WiFi. Following are the contributions

made by WiFi-Nano-

Eliminated SIFS by speculative ack preamble transmission and thus reduced the ack

overhead.

Parallel implementation of preamble detection with preamble transmission. This

benefits in collision reduction and unfairness elimination.

Introduce shorter slot time of 800 ns and in turn improves the throughput of WiFi by

up to 100%.

Channel access and ack overhead was reduced.

Introduced the technique for sub-preamble detection and its realization by using a

lattice correlator.

Improved the air-time efficiency (fraction of time data was transmitted over the air)

of WiFi.

1.3 Methodology

Following methods were adopted to increase efficiency in WiFi-Nano-

800 ns slots : Instead of using 9 microseconds slots as in 802.11, 800 ns slots were

used in WiFi-Nano. It makes back-off efficient.

Speculative Preamble Transmission: Instead of waiting for multiple slots for detecting

preambles, nodes in Wifi-Nano speculatively transmit preambles as their back-off

counters expire, while continuing to detect preambles using self-interference

cancellation. Contention for channel access is carried out simultaneously with

preamble transmission. All the devices abort their transmissions midway except those

whose back-off counters expired the earliest. Thus average channel access time can

be reduced to 7.6 microseconds which was 101.5 microseconds with 9 microseconds

slot time and 100 bytes of packet size in 802.11 Wifi.

Speculative ACK: Instead of waiting for SIFS before transmitting the ACK preamble,

the receiver speculatively starts transmitting its ACK preamble as soon as it finishes

reception of the packet. The ACK transmission is then aborted midway upon detecting

errors in the received packet. Thus Speculative ACK Transmission allows WiFi-Nano to

eliminate SIFS and thus reduce the ACK overhead.

Working with Lattice Correlator: A novel Lattice Correlator was designed to enable

chained contention resolution in WiFi-Nano. In order to do this, their correlator must

provide two functions. First, devices are required to correlate sub-parts of a preamble.

Second, the need for roll-back requires that the exact position of the correlation be

known, since this will help accurately determine the beginning of the packet

transmission. Each packet of WiFi-Nano is preceded by a PN sequence comprising

several short but distinct 800ns PN sequences PN1, PN2, ···, PNn. The lattice correlator

takes as input the received signal, and generates , (N is the number of 800ns PN

sequences) outputs corresponding to the correlations obtained from each continuous

sub-part of the preamble e.g., [PN1, PN2], [PN3, PN4, PN5] etc. Detection of a spike

in any of these inputs provides two pieces of information. First, the presence of an

ongoing transmission, and second, the start time of the beginning of the reception.

The start time of beginning of the packet reception is determined by the position of

the last 800ns PN sequence.

Probabilistic Collision Resolution: Since potential collisions can be detected in each

800ns slot, WiFi-Nano uses a novel contention resolution scheme to resolve collisions

on the fly. Finally, note that when more than two packets collide in a given slot, the

number of collisions can be approximately estimated by the number of correlation

spikes that occur within a single 800ns slot (this is because the slot boundaries of

different nodes are not perfectly aligned due to differences in propagation delays).

Upon detecting k − 1 distinct spikes in a single slot, rather than using 50%, each device

continues transmitting with a probability of . Thus, the probabilistic collision

resolution mechanism in WiFi-Nano avoids payload collisions with a high probability,

thereby significantly reducing the collision overhead

1.4 Conclusion

The main objective of this research was to examine the efficiency by reducing the 9

microseconds slot used in Wifi to 800 ns. The 800 ns slot size needs some alteration in the

conventional 802.11 Wifi like the preamble transmission and detection was done in parallel,

which is achieved by using speculative transmission of preambles and analogue interference

cancellation. Also SIFS is eliminated by speculative ACK transmission. Furthermore, a novel

lattice correlator is designed that correlates to parts of the preamble and is able to accurately

determine the start time of detected preambles, which is a key requirement for accurate

rollback of speculative preamble transmissions. Also, Nodes in Wifi-Nano abort transmissions

probabilistically since nodes are able to detect collisions during the preamble transmission

phase. This way, the packet collisions are avoided with high probability.

2. Critique

2.1 Lack of information about DIFS

The paper talked about the elimination of SIFS in WiFi-Nano by speculative ACK preamble

transmission but had not talked about DIFS time. Whether, DIFS is there in WiFi-Nano or it is

also eliminated, is also not clear in the paper. Since the slot time is reduced to 800 ns in WiFi-

Nano, there must be an impact on DIFS as well if it is there. Generally DIFS is equal to (SIFS +

2 * slot time). Since SIFS in WiFi-Nano has been eliminated, so DIFS should be equal to 2*800

i.e. 1600 ns if it is present. Also they haven’t talk much about the size of contention window.

What should be the size of contention window so that both fairness and efficiency of WiFi

can be maximized?

2.2 Challenges in speculative preamble transmission

Nodes may unduly count down their back-offs while other nodes’ preamble are being

transmitted. This would result in some wastage of resources. The preamble may be detected

as late as 5 slots after the preamble transmission, the channel is perceived as free. Due to

this, there is some overhead in efficiency also. This problem is not discussed in the paper.

2.3 Why 800 ns slot?

Why the author has chosen this figure of 800 ns as the slot time? The author surely want

to decrease the overheads by reducing the slot time. Does the overhead linearly dependent

on slot time or not? There must be some relation between the slot time and the overheads.

The author neither mention why he picked 800 ns slot nor he discussed the relation between

slot time and overheads.

3. Synthesis

Wifi-Nano is designed to increase the efficiency of WiFi with high data rates. The

technology can be tested on smaller data rate WiFi (11 Mbps or 54 Mbps) and study the

results. Since these WiFis already have greater efficiency, so it might not greater impact on

these but even a 10% improvement in efficiency would be great if it will be able to do that.

There are various techniques discussed in the paper for solving some problems like

collision overhead, fairness etc. These solutions can be explicitly used in other scenarios, for

example, we can use Probabilistic collision resolution in 802.11 or other variants of WiFi. We

can use the novel lattice correlator designed for Wifi-Nano in other purposes as well.

The paper has presented four techniques to improve the efficiency of WiFi. They are

speculative preamble transmission, probabilistic collision resolution, counter roll back and

speculative ACK transmission. All the four techniques are very specific and can be used

explicitly for other purposes.

Likewise the SIFS is eliminated from the WiFi, one more extension could be to eliminate

the DIFS time as well. There may be some way possible to eliminate it as well.