Friday, August 12, 2011

Turn down that NOISE!!!

Here is a fun experiment if you have a 2 laptops (or 1 laptop and some other WiFi enabled device), a WLAN analyzer like AirMagnet WiFi Analyzer (or something similar such as wireshark) and a microwave oven.

We all know that microwave ovens function in the 2.4Ghz range but how much "noise" is it generating.

Go to your kitchen. Get the device without an analyzer on it and start a really long file transfer via WiFi. Make sure your AP is already on either channel 11. Now Fire up your analyzer and measure the signal, noise and the signal to noise ratio (SnR is signal minus noise) on channel 11. Get an average reading and write it down. Do it several times so you are sure you have a good sample set. Now, restart your file transfer, put a pyrex (or other microwave safe) measuring cup full of water in your microwave and set it for a few minutes on high. Re-measure your signal, noise and SnR. What did you get? Try a few different WiFi adapters with the analyzer. any difference?

Here is what I saw when I did it with the AirMagnet WiFi Analyzer. First is the image of the channel screen on channel 11 with no microwave turned on. Notice the stats.


Signal:      -51dBm
Throughput:  ~34Mb/s
Utilization: 68%
Noise:       -88dBm
SnR:         37 (-51) - (-88) = 37


Here is the important part. Notice that the noise level is completely flat at -88dBm.

Now lets turn the microwave on and do it again, here is what I saw.


Signal:      -47dBm
Throughput:  ~16Mb/s
Utilization: 39%
Noise:       -88dBm
SnR:         41 (-47) - (-88) = 41

Notice anything? Well first off throughput went down and I checked the capture later and packet loss went up. Seems appropriate with a 1200watt device slamming my connection from 2 feet away. Secondly, the signal strength and the SnR INCREASED? Why is that? Well after looking deeper into the frames and looking at the AP configuration, it looks like 802.11N beamforming in action! The AP senses packet loss and pushes harder (no-tech speak for adjusts the phase shift of the signal to focus the waves on top of the client so it gets the benefit of wave addition). Next thing you notice is the increase in SnR from 37 to 41. This is due to two issues. The first we just mentioned. Increased signal strength. The second is the trouble spot. The noise level stayed the same.

What?!!

You heard right, the noise level did not change. I can hear you yelling at me through the intertubes from here. But I thought microwave oven caused noise? Well. they do and they do not. Here I am about to get pretty picky.

You see we at AirMagnet call noise one thing and interference another and WiFi interference a third. Confusing but true. And with good reason. You see, noise readings are fiction. They are falsehood, fabrication, fib. It is a lie. There. I said it. Noise readings (except from a very small number of cards - namely only one kind that I know of) are made up on the spot by each WiFi adapter. They cannot give you a noise reading, they can only give you bits (e.i. 1s and 0s) about things they see and WiFi adpaters only see WiFi modulation. Period. Almost all wifi adapters throw away all other non-WiFi signal long before it ever gets to the the driver. Non-wifi modulation, such as the modulation you get from cordless phone, wireless headsets, video camers etc are thrown away by the radio long before they can be converted to bits. The radios are only tuned for the type of modulation and coding they understand. So a normal 802.11a adapter will only ever pass OFDM modulation. Never QPSK, or FHSS or any other signal that is not OFDM. Thus how can the driver tell us about a microwave oven, or a cordless phone, wireless camera etc? They can't. So we call interference from other modulation types in the same frequency band by a different name, we call it non-802.11 interference.

Then there is WLAN interference, which Tom's Hardware calls "congestion", but could also be thought of as WLAN contention and is sometimes referred to as co-channel interference or even adjacent channel interference. Tom's has a very good article on it here. It is where too many WLAN devices are all trying to use the same medium simutaneously. In 802.11 terms, only one device can talk in an area at one time. Too many devices means collisions and contention in a collision avoidance network that wishes to remain contention free.

So that still leaves us with noise. What is going here? When I open a WLAN analyzer I see a noise reading. You showed me one right there (see images above). Where does that come from?

To understand what the adapter manufacturers are doing requires that we first understand what noise is. Noise can be thought of as the level at which all radio signal become indistinguishable from one another. Wikipedia describes the noise floor as, "the measure of the signal created from the sum of all the noise sources and unwanted signals within a measurement system." You can create an analogy of this as follows: Imagine five people in a room, each yelling in a different language. You could probably figure out what each language was even if you couldn't have a conversation in there, you might even be able to work out what they are talking about. That example would be considered interference to you if you were trying to yell in English. Now imagine a room with 500 people in it all talking at the same time. in that instance you proabaly could not tell what any of the languages were or what anyone was talking about at all. That is noise. Again, the Tom's Hardware article is a great reference for this, especially slide 9.

 Adapter manufacturers were asked many times for a noise reading so the person conducting the WLAN site survey could determine the signal to noise ratio, or SnR. This helps determine if there is enough signal to be heard above the RF noise floor. So under this presure they created algorythims that can fake the noise reading. I have no idea what the algorythims are but I can assume from tests we have done at AirMagnet that they include measurements of retry rates, CRC errors, channel utilization, average throughput and, mostimportantly, number of devices active in range (as determined by RSSI). The reason I can know this is because at AIrMagnet we have an RF isolation chamber - or Faraday Cage. If I take a noise reading in that chamber with one AP and no STAs (WLAN client adapters) I get a zero reading with many adapters. But if I add STAs the noise floor rises. even if the STAs are mostly dormant.

Here is another catch, aside from the fact that different manufactureres create this value differently, some adapters only give noise readings on a channel by channel basis, others frame by frame and still others give no reading at all. Symbol devices fit this last group and when asked about why they gave no noise reading they replied, "that would be lying". Does that mean that you cannot use that value to create the much sought after SnR? No, I would still use it, but I would be wary if tehre were very few STAs nearby. Or many for that matter.

I hope this helps understand what noise is, what interference is and what co-channel interference or WLAN contention is.

6 comments:

  1. Amen Brother!

    I've been preaching NOT to trust the 'Noise Variable' for nearly a decade. Thanks for writing a blog to help more folks understand.

    Keith

    ReplyDelete
  2. Great post, Bruce. Thanks for sharing.

    ReplyDelete
  3. Great post. The thing about wireless cards reporting non-wifi interference is not true --- newer atheros cards (e.g., AR9280) provide the functionality to get FFTs which can be used to measure non-WiFi interference. People at UW Madison have done this; They were able to detect non-WiFi devices like microwaves, analog phones, zigbee, cordless phones and bluetooth devices --- all using a WiFi card!!

    ReplyDelete
  4. Agreed. The latest gen Atheros cards do provide this function and previous gen Intel cards also did. Unfortunately, they represent a very small minority of tools available to people today and a very many are still thinking Noise is Interference. I hope that eventually all adapters will provide a true FFT interference score as it only makes sense.

    -BH

    ReplyDelete
  5. I see that they now have the demo of a WiFi card being able to detect Bluetooth, Cordless phones, Microwaves, Xbox, Videocams
    http://www.youtube.com/watch?v=VuAzXhcA8h8 and the paper here: goo.gl/AwNSN

    ReplyDelete