도닦는공돌이

Ways to increase Wi-Fi performance. Part One: Bursting, Compression, 

Fast Frames, Concatenation
Read more at http://ixbtlabs.com/articles2/comm/tech-80211g-super.html#8G5M62mYj7e8urzI.99

Practically all presently manufactured 802.11g wireless adapters have such suffixes as "super G", "turbo", "plus", etc. But suffixes are only half the work. Manufacturers (to be more exact, their marketing specialists) decorate their boxes with the 108 Mbit/s or even 125 Mbit/s labels.

125 - sounds tempting. Can it really be true that wireless adapters work faster than the good old Fast Ethernet with its cables? Maybe we should let them go... those "ancient" Fast Ethernet adapters? Get rid of cables, we are sick and tired of, and long live Radio Ethernet? :)

But look before you leap, as the proverb runs. In our case it means that it would do no hurt to find out more details about these mysterious techniques, how they work, and what data rates they really provide (and the most important thing - under what conditions). In other words, we should make allowance for the most important thing to marketing guys - to sell solutions from their company.

There are many ways "to overclock" the standard 802.11g. To be more exact, every chip manufacturer has its own way (at least - they are called differently). Unfortunately, not all manufacturers explain the details of their techniques. I managed to find information on these techniques only for Atheros and Texas Instruments. But the most informative resource is provided by Atheros - it even has a separate web site, devoted to their Super G and Super AG techniques.

In fact, most part of this article is a compilation of the information from the web sites of Atheros and Texas Instruments and only minute details from other sources.

Let's proceed to the techniques.

At first let's have a look at the "pure" 802.11g. Maximum throughput of this mode is 54 Mbit/s. I guess the majority of users know how to convert megabits into megabytes? That's right, you should divide megabits by eight to get the data rate - 6.75 MB/s.

But attentive readers (those who don't just look through introductions and conclusions, but actually browse diagrams with performance readings) know that the regular 802.11g mode does not provide more than ~25 Mbit. Hey, that's only half of the 54 Mbit! Where is the other half? "Where" is a topic in its own right. I can only note that, indeed, user data makes for only half (at best) of the channel bandwidth.

That's the first bad news. There is also the second bad news. Radio waves (it's actually them who transmit data in wireless networks) are transmitted in all directions from the signal source (it's a general case). That is everyone hears the transmitter. Everyone can choose to receive data or not, that's not important. What's important - these people cannot transmit anything on the same frequency at that moment. To be more exact, they can try, but signals from both sources will overlap, which will lead to the distortion and loss of the data. In other words, only one of several sources operating at the same frequency can transmit data at a time in wireless networks. That is the walkie-talkie principle - first you talk, then you keep silence and listen.

Thus, the generously allocated ~25 Mbit are divided between all participants of a wireless network. If the number of clients is 5 hosts and all of them are actively transmitting data, every participant will have the bandwidth of about 5 Mbit (in fact, it will be even a tad lower).

There is also the third bad news. The second bad news about 5 Mbit per each of 5 hosts is true only in case of Ad Hoc network, that is without an access point. If we take a more general case with an access point, those miserable 5 Mbit will have to be divided into two. Any exchange with clients in the Infrastructure mode (with an access point) goes via an access point. At first it should receive data and then retranslate it to the recipient. As a result, we get 2 and a little more megabit per user.

Now let's get back to the figures 108 and 125, which are often printed in large font on product boxes. But you have already got it, right? :)

Divide these figures by two (we'll touch upon the ideal case later). That means 60 Mbit maximum in case of one client and consequently n-times smaller in case of N clients.

If all you want is to find out whether it's time to get rid of wires or "wait a little", you may skip the rest of the article. The answer is it's too early so far. At least wait for WiMAX.

Now let's proceed to a more thorough examination of techniques for increasing wireless throughput compared to the standard 802.11g mode.

I guess all advantages (turbo, etc) of the manufacturers are just the same thing as in TI and Atheros, but under different names. But implementation details may be different, so techniques from different manufacturers may be incompatible with each other.

Atheros technique for 802.11g is called Super G (there is another one - Super AG; it's the same thing, but for 802.11a, i.e. for 5 GHz networks). Atheros Super G allows to increase the throughput to 108 Mbit/s. As Atheros honestly declares, user's data rate may reach 60 Mbit.

Performance is increased by several methods:

Atheros Super G / Super AG techniques:

Atheros web site contains a colorful diagram that shows the influence of various techniques on data transfer rates:

Pic.1, Benefits of various techniques for wireless performance

The base 802.11g or 802.11a mode, where all extended techniques are disabled, allows up to 22 Mbit (net value, that is available to a user). Adding techniques, which will probably be included into the future 802.11e standard (Bursting, Fast Frames, Compression), we can increase the performance up to 40 Mbit inclusive. Activating Dynamic Turbo mode, that is utilizing dual channels for data transfers, may increase performance to the theoretical maximum of 60 Mbit.

The above figures are certainly the maximum possible performance in a given mode (in the ideal case). In reality everything will depend on such conditions as client's distance from an access point, a number of clients operating simultaneously, radio environment around the wireless network, etc.

Wireless performance boosts from Texas Instruments are called G-Plus. Some of them resemble techniques from Atheros, the others are characteristic of TI alone.

Texas Instruments G-Plus techniques:

Let's dwell on each of the above mentioned techniques - bursting, compression, fast frames, dynamic turbo. Interestingly, all the four techniques work independently, thus maximizing performance simultaneously in several ways.

1. Bursting

Frame Bursting is a transmission technique supported by the draft 802.11e QoS specification. Frame Bursting increases the throughput of any (point-to-point) 802.11a, b or g link by reducing the overhead associated with the wireless transmission. This results in the ability to support higher data throughput in both homogeneous and mixed networks.

Picture 2 shows an example of a standard transmission (without bursting).

Pic.2, Standard 802.11a/b/g mode

In standard mode we can see the process of transmitting two frames (frame1 and frame2) from Source to Destination in time. The process of data transmission is divided into time intervals (axis X is time). As only one source can transmit data at a time, each station should contend for airtime during DIFS (Distributed InterFrame Space). If no other station is transmitting, the airtime is free and a frame can be transmitted. After a frame is transmitted (frame1), the transmitter waits for a confirmation on a successful delivery from the destination. The destination must send an acknowledgement (ack) practically immediately after SIFS - Short InterFrame Space (if there was no acknowledgement, the source considers that the frame was not received and must resend it). After receiving an acknowledgement, the sender must again wait for DIFS and only then (if the air is still free) start sending Frame 2. And so on.

Thus, DIFS take up a considerable part of wireless throughput.

Now let's see the picture of Frame Bursting transfer:

Pic.3, With Frame Bursting

In this mode (Picture 3), the source and the destination capture a channel in turns for their transmissions. After frame1 is transmitted and the acknowledgement is received, the transmitter does not wait the required DIFS. The sender waits only SIFS and then transmits the second data frame, etc. Thus, the sender does not give an opportunity for other stations to start transmissions - they have to wait for the end of this burst transmission.

Of course, the total transmission time in this mode is limited (otherwise, transmission of several gigabytes would have paralyzed other clients of a given wireless network completely). But eliminating DIFS allows a larger chunk of data transmitted over the same period of time, thus saving the channel throughput, that is increasing the total transmission performance.

Atheros announces that all its products support this technique. But devices from other manufacturers, which do not support this technique, may fail to understand this burst mode. So, if communicating with a product that fails to acknowledge a burst transmission, the source falls back to the base mode.

Implementation of Bursting from TI is similar to Atheros. TI provides the following illustration of its technique (Picture 4):

Pic.4, Frame Bursting from Texas Instruments

TI also eliminates the "long" interframe space, by reducing overheads on transmission.

Both companies do not provide information on compatibility of burst techniques from TI and Atheros.

Similar "bursting" techniques are probably offered by other manufacturers as well. But Atheros went further than that and expanded this technique to "dynamic bursting". It announces that this technique is especially effective in networks with the number of wireless clients exceeding one.

For example, if there are two stations, near and far from an access point. Of course, the far station will operate with an access point at a lower data rate (because of the distance). That's why its transmission (to the near client) of a given size will take more time than it will take the near client to receive the data. In this case bursting activation for the far station allows to reduce the airtime and, strange as it may seem, it also allows the nearby station to receive this data still faster (since it will spend less time contending for airtime). Burst transmission periods also depend on the distance (to be more exact, on data rates). The nearby client is granted a longer burst transmission, as it will burst more frames, while consuming much less airtime.

Atheros Compression technique

The second technique from Atheros that extends the 802.11 standard is hardware compression. It's built into all 802.11a,b,g chipsets from this company. It uses the Lempel Ziv algorithm. The same algorithm is used in such archivers as gzip, pkzip, winzip. This engine compresses prior to transmission and decompresses after reception.

Unfortunately, the data is not analyzed before it's compressed, all frames are compressed. Thus, it's not always good - for example, sending an already compressed file may increase the size of wireless transmission.

On the other hand, well compressible data will be transmitted in smaller frames, thus consuming less airtime. This airtime can be used by other wireless stations.

Atheros Fast Frames

Fast Frames technique bundles two frames into a single larger frame. Thus we eliminate extra overheads (in the header of the second packet - there is only one header in the new packet left) and interframe spaces:

Pic.5, Regular Transmission

Pic.6, with Fast Frames

The size of the resulting frame may reach 3000 bytes, which is twice as large as the maximum frame size of a standard ethernet packet. Thus, Fast Frames technique will work even with wireline transmissions with maximum packet size (1500 bytes), by merging each two ethernet packets into a single larger packet. Once FastFrames have been negotiated between an access point and a station, both the AP and the station can send wireless frames of 3000 bytes to the corresponding peer.

Considering that Fast Frames can operate together with Frame Bursting, we get very good data rates. By the way, according to Atheros, most manufacturers, using Frame Bursting in their chips, actually don't support Fast Frames. Atheros is all right - their devices support both techniques.

The Fast Frames technique is also based on the 802.11e draft standard. Nevertheless, it may not be supported by all third party hardware. On the other hand, this technique functions within existing timing parameters (unlike Frame Bursting, which exclusively captures a channel for some time). That's the reason Fast Frames are better for wireless networks that use equipment from various manufacturers.

Texas Instruments Frame Concatenation

The Frame Concatenation technique in devices from Texas Instruments uses the same principles as Fast Frames from Atheros.

TI goes further than that. In this case two or more frames are merged (Picture 7):

Pic.7, Frame Concatenation

Thus, it gains by eliminating overheads and interframe spaces from one or more frames. TI claims that its Frame Concatenation technique will work with any 802.11b/b+/g devices from TI and (!)other manufacturers. It's not quite clear what this company meant by other manufacturers, if the latter don't support this technique... Perhaps it meant operations with frames that don't exceed the standard size (1500 bytes).

Frame Concatenation incorporates an algorithm that allows to merge only selected packets into mega-frames. For example, if there is only one frame in queue to be sent to a given destination, it will be sent immediately. In other words, only frames to the same destination address (MAC address of a receiver, in this case) will be merged. The algorithm works only with unicast packets - multicast packets as well as control packets are sent without changes.

At present, the maximum size of a Concatenation packet may reach 4096 bytes (which is an indirect sign that this technique is not compatible with its Atheros counterpart).

Conclusion

As we can see, manufacturers are not waiting for the official announcement of the standards (802.11e in this case), but integrate these new techniques into their products. Thus, they obtain good results as far as performance gains are concerned, on the one hand. But on the other hand, techniques from different manufacturers are often incompatible with each other.

We haven't reviewed Dynamic Turbo from Atheros yet. It will be described in the second part of the article.

If we find documentation about such techniques as super/plus/etc from other manufacturers of wireless solutions (or if you post links to such documents in our forum (the link to our forum is right after this article, a tad below)), reviews of these techniques will also be added to the second part of the article.

Write a comment below. No registration needed!

Read more at http://ixbtlabs.com/articles2/comm/tech-80211g-super.html#8G5M62mYj7e8urzI.99

결론: Wi-Fi에서 Burst mode를 사용하는 모듈(Atheros,TI..)을 구매하여 사용해야 하며,  이는 Burst mode 모듈이 Backoff를 수행하지 않고 SIFS만큼만 대기후 연속적으로 데이터를 전송하기 때문에, 다른 기기들의 채널접근을 막는다. 따라서, 단기간만 Throughput을 올리는 목적으로 사용해야한다. 

* I sincerely promise that this paper would be used for purely academic purposes only and not for any commercial applications. this copyright belong to original author.  - from blog owner.

This content refer to "Channel Models A tutorial' paper of Raj Jain, jain@acm.org author in Google.

V1.0February 21, 2007

Channel Models A tutorial

1. The Basic Concepts

Wireless signal의 특성은 그것이 전파하는 (송수신간의) 매체를 통과하면서 변하게 된다.
이러한 특성은 송,수신 안테나 사이의 거리에 의하여 의존하게 되며, 그 거리에는 신호가 지나가는 경로와 환경(방해가능한 빌딩과 어떠한 물체 -> 이는 신호의 반사 및 흡수가 가능하다.) 적인 요소가 존재하여 채널의 특성을 변화시킨다.

만약 우리가 송수신 안테나사이의 매체 i.e 채널을 알수있다면, 입력신호를 앎으로써, 출력신호를 알수있게 된다.

Therefore, this model of the medium is called channel model.

In general, the power profile of the received signal can be obtained by convolving the power profile of

the transmitted signal with the impulse response of the channel. Convolution in time domain is

equivalent to multiplication in the frequency domain. Therefore, the transmitted signal x, after

propagation through the channel H becomes y

y(f)=H(f)x(f)+n(f)

Note that x, y, H, and n are all functions of the signal frequency f.

The three key components of the channel response are path loss, shadowing, and multipath as explained below.

따라서, 우리는 채널에 영향을 주는 대략적인 3가지의 성분 path loss, shadowing, and multipath 를 통하여서, 채널을 추정할수가 있게 된다.

2.1 Path Loss

The simplest channel is the free space line of sight channel with no objects between the receiver and the transmitter or around the path between them. In this simple case, the transmitted signal attenuates since the energy is spread spherically around the transmitting antenna. For this line of sight (LOS) channel, the received power is given by:

먼저, 가장 간단한 형태의 채널을 정의해보자. 이는 자유공간 즉, 벽면도 없고 반사도 없는 그래서 손실도 없는 LOS (direct path propagation) 타입이다. 

감쇠의 원인은 안테나의 전파(신호) 방사형태가 구면의 형태로 에너지를 확산시키기 때문에 P는 D^-2에 비례한다.

Here, Pt is the transmitted power, Gl is the product of the transmit and receive antenna field radiation patterns, λ is the wavelength, and d is the distance. Theoretically, the power falls off in proportion to the square of the distance. In practice, the power falls off more quickly, typically 3rd or 4th power of distance.

G1은 안테나의 방사패턴에 해당하는 Loss를 의미한다.

이론적으로 파워는 거리의 제곱에 비례하여 감소하지만, 사실 실제환경에서는 이보다 빠른 세제곱, 네제곱배로 감소한다.

The presence of ground causes some of the waves to reflect and reach the transmitter. These reflected

waves may sometime have a phase shift of 180 ° and so may reduce the net received power. A simple two-ray approximation for path loss can be shown to be:

자 이제 Free space 시나리오를 좀더 꼬아서 Ground 가 존재한다고 가정한다면, 이는 반사를 야기시킨다. -> 위상을 180도로 변화시키며, Power를 감소시킨다.

Here, ht and hr are the antenna heights of the transmitter and receiver, respectively. 

Note that there are three major differences from the previous formula. First, the antenna heights have effect. Second, the wavelength is absent and third the exponent on the distance is 4. 

이 공식의 앞전의 위의 공식과 3가지의 주요한 다른점이 존재한다.

안테나높이 고려, 파장고려X, 지수부4승.

In general, a common empirical formula for path loss is:

그리고 이를 더욱 일반화한 형태로, 경험에 의한(실험에 의한) 공식은 아래와 같다.

위의 공식은 실제적인 path loss를 고려한 형태의 수신되는 신호의 Power를 의미한다.

Where P0 is the power at a distance 0 d and α is the path loss exponent.

The path loss is given by:

Path Loss 공식은 아래와 같이 dB표현식으로 주어진다

Here PL(d0 ) is the mean path loss in dB at distance 0 d . The thick dotted line in Figure A.1.2 shows the received power as a function of the distance from the transmitter.

If you want to know the more detail information about this section, please refer to Fundamentals Wireless Communication text book's chapter 2. It's free on website. 

http://www.eecs.berkeley.edu/~dtse/book.html

2.2 Shadowing

If there are any objects (such buildings or trees) along the path of the signal, some part of the transmitted signal is lost through absorption, reflection, scattering, and diffraction. This effect is called shadowing. As shown in Figure A.1.3, if the base antenna were a light source, the middlebuilding would cast a shadow on the subscriber antenna. Hence, the name shadowing.

쉐도잉은 송신 신호가 수신단으로 가기까지 어떠한 장애물이 존재하여, 송신신호가 흡수,반사,회절,흩어짐을 겪게 됨으로 소실되거나 딜레이되어 수신단에 도착하게 되는데 이로 인해 다른 크기와 위상의 신호가 중첩되어서, 이를 평균전력그래프로 그려보았을 때, "로그노말분포"로 나타나기에 흔히 우리는 로그노말쉐도잉이라고 부른다.

[보충]

Shadowing은 문자적 뜻은 음영 손실 (Shadowing Loss)을 의미하며, 현상은 전파 장애물 바로 뒤에 전파적인 그림자(음영)가 드리워져 나타나는 전파 손실을 말한다. 예를들면, 송신 전파가 산,빌딩 등 장애물의 불규칙적인 전파 반사면,산란체 등 때문에 수신 전파 전력이 평균을 중심으로 요동치는 현상 이라고 볼 수있다. 

The net path loss becomes:

Here χ is a normally (Gaussian) distributed random variable (in dB) with standard deviation σ . χ represents the effect of shadowing. As a result of shadowing, power received at the points that are at the same distance d from the transmitter may be different and have a log-normal distribution. This phenomenon is referred to as log-normal shadowing.

2.3 Multi-path

The objects located around the path of the wireless signal reflect the signal. Some of these reflected waves are also received at the receiver. Since each of these reflected signals takes a different path, it has a different amplitude and phase. 

반사되는 신호들이 서로 다른 경로를 가지므로, 다른 위상,크기를 가진 신호가 되기에 이를 다중경로 페이딩이라고 부른다.

[보충] : Multi-path Fading

서로 다른 경로를 따라 수신된 전파들이 여러 물체에 의한 다중 반사로 인해 서로다른 진폭,위상,입사각,편파 등이 간섭

(보강간섭,소멸간섭)을 일으켜 불규칙 요동치는 현상

Depending upon the phase, these multiple signals may result in increased or decreased received power at the receiver. Even a slight change in position may result in a significant difference in phases of the signals and so in the total received power. The three components of the channel response are shown clearly in Figure A.1.4. The thick dashed line represents the path loss. The log-normal shadowing changes the total loss to that shown by the thin dashed line. The multipath finally results in variations shown by the solid thick line. Note that signal strength variations due to multipath change at distances in the range of the signal wavelength.

그래서 아래 그림과 같은 형태의 수신 파워를 보인다. 아래의 그래프는 x축 거리가 멀어짐에 따른 y축 수신파워/송신파워의 db비를 나타낸 그래프이다. 이를 통하여 각 채널에 영향을 주는 3가지 성분들로 인하여 입력신호가 어떻게 수신되는지를 대략적으로 알수가 있다. 

Since different paths are of different lengths, a single impulse sent from the transmitter will result in multiple copies being received at different times as shown in Figure A.1.5

아래의 그림은 송신 신호가 수신측에서 Multipath를 겪고서 어떻게 페이딩현상으로 나타나는지를 보여준다.

(즉,LTI System으로 수신신호를 입력신호대 채널간의 컨벌루션으로 모델링한 형태를 떠올리면 이해가 쉽다)

The maximum delay after which the received signal becomes negligible is called maximum delay spread τmax. A large τmax indicates a highly dispersive channel. Often root-mean-square (rms) value of the delay-spread τrms is used instead of the maximum.

송순 후에 multipath fading으로 너무 많은 딜레이와 감쇄효과가 있어서 무시해도 되는 수신신호파워를 우리는 τmax 라고 부른다.

3. Tapped Delay Line Model

One way to represent the impulse response of a multipath channel is by a discrete number of impulses as follows:

Note that the impulse response h varies with time t. The coefficients ci(t) vary with time. There are N coefficients in the above model. The selection of the N and delay values τi depends upon what is considered a significant level. This model represents the channel by a delay line with N taps. For example, the channel shown in Figure A.1.5 can be represented by a 4-tap model as shown in Figure A.1.6.

멀티 패스 채널의 임펄스 응답의 표현은 위의 식처럼, 임펄스들의 이산적인 합으로 표현될수가 있다.

[보충] : 왜 임펄스 함수인가?

흔히 채널은 시스템으로 표현할수가 있다고 한다. 왜냐하면, 입력신호를 어떤 시스템에 넣고 (채널) 출력신호를 보는 과정이 통신의 송/수신과 매우 흡사하며, 이를 수학적으로 모델링하여 나타내면 간단 명료하게 채널의 특성을 보여줄수 있기 때문이다. 임펄스 응답이란? 시스템에 단위 임펄스함수를 인가하였을 때의 출력을 의미하며, 임펄스 응답에 대한 '푸리에변환'은 통신의 주파수 특성 (전달함수 H)와 같다. 그래서 임펄스 응답을 알면 우리는 주어진 입력에 대한 출력을 쉽게 구할수 있기 때문에 우리는 임펄스 함수를 사용하게 되었다.

If the transmitter, receiver, or even the other objects in the channel move, the channel characteristics change. The time for which the channel characteristics can be assumed to be constant is called coherence time. This is a simplistic definition in the sense that exact measurement of coherence time requires using the auto-correlation function.

만약 송/수신기(ex.안테나)가 움직이면 채널의 특성도 변하게 된다. 그래서 우리는 채널이 제한된 특성을 가지고 있는 시간을 하나의 척도로 명명하였는데 이것이 coherence time 이다.

coherence bandwidth 도 이와 매우 유사하다.

For every phenomenon in the time domain, there is a corresponding phenomenon in the frequency domain. If we look at the Fourier transform of the power delay profile, we can obtain the frequency dependence of the channel characteristics. The frequency bandwidth for which the channel characteristics remain similar is called coherence bandwidth. Again, a more strict definition requires determining the auto-correlation of the channel characteristics. The coherence bandwidth is inversely related to the delay spread. The larger the delay spread, less is the coherence bandwidth and the channel is said to become more frequency selective.

delay spread는 LOS와 비교하였을 때, 위의 Figure A.1.6과 같이 지연되어 수신되는 현상을 말한다. 

delay spread가 크다면 coherence bandwidth 은 작으며, 이때 채널을 우리는 frequency selective하다고 말한다.

4. Doppler Spread

The power delay profile gives the statistical power distribution of the channel over time for a signal transmitted for just an instant. Similarly, Doppler power spectrum gives the statistical power distribution of the channel for a signal transmitted at just one frequency f. While the power delay profile is caused by multipath, the Doppler spectrum is caused by motion of the intermediate objects in the channel. The Doppler power spectrum is nonzero for (f-fD, f+fD), where fD is the maximum Doppler spread or Doppler spread.

The coherence time and Doppler spread are inversely related:

coherence time과 Doppler Spread는 서로 역수관계를 가진다.

Thus, if the transmitter, receiver, or the intermediate objects move very fast, the Doppler spread is large and the coherence time is small, i.e., the channel changes fast.

Table A.1.1 lists typical values for the Doppler spread and associated channel coherence time for two WiMAX frequency bands. Note that at high mobility, the channel changes 500 times per second, requiring good channel estimation algorithms

높은 대역으로 Carrier Frequency가 갈수록 Coherence Time은 작으며, Max Doppler Spread가 높을수록 채널의 안정적인 시간 = 주파수가 Flat한 시간이 적다.

여기에선 Coherence Time과 Doppler Spread만을 비교하였으며, Doppler Spread에 대한 자세한 정보가 필요하다면 위에서 소개한 책의 Chapter 2.1.4를 참고하기 바란다.

in this Tutorial, we just handled the comparison with Coherence Time and Doppler Spread. If you want to learn the Doppler Spread, you can easier understand though the textbook which we already introduced.