top of page

Follow me on Twitter or RSS to get notified when a new blog is released.

  • Twitter
  • LinkedIn Social Icon
James Scott

Non-overlapping channel interference

Updated: Aug 22, 2019

After speaking with several people concerning WLAN interference, I've come to realize that many people are not aware of interference caused by the spectral mask guidelines from the 802.11-2016 standard. WLAN interference is an issue that will cause you headaches and grief, if you don't understand the underlying cause.


In figure 1 below, you will see the spectral mask requirements set by the IEEE. This image and the respective description of it can be found in the 802.11-2016 standard in section 21.3.17.1 on pages 2582-2585.


We will concentrate mainly on the 20 MHz spectral mask. And we will take a brief look at a 80 MHz channel. Just know that the wider the channel, the more potential for interference there is. And remember that antenna aperture matters, the 2.4 GHz band will suffer from these effects more than the 5 GHz band, because of antenna aperture.

Figure 1

Figure 1 is showing us the main signal lobe is contained between -9 and +9 MHz (this is for all current amendments that use OFDM). Inside this area, there is a 0 dBr (dB relative to the maximum spectral density of the signal). Meaning there are no dB adjustments for the -9 to +9 MHz range. Between -11 to -9 MHz and +9 to +11 MHz there must be at least -20 dBr, between -20 to -11 MHz and +11 to 20 MHz there must be at least -28 dBr, and between -30 to -20 MHz and 20 to 30 MHz there must be at least -40 dBr.


Lets take a look at what FSPL has to say about this, and then we will look at some real life examples.


It's recommended that you keep any AP at least 10 ft from another AP, even on different channels, unless they are separated by at least one non-overlapping channel, such as using channels 36 and 44 in 5 GHz, or channels 1 and 11 in 2.4 GHz. This is according to at least one Vendor: Aerohive, and according to FSPL Otherwise you risk causing some ACI, from the CCA ED mechanism. Remember that the CCA requirements for OFDM are in the 802.11-2016 standard in section 17.3.10.6, and state that energy detect will trigger a medium busy if any signal of -62 dBm or greater is sensed and will be triggered busy for any preamble / PLCP header, or frame at -82 dBm.


The noise at lower power levels can be the real problem though. You are going to increase the noise floor, and decrease SNR when two APs are stacked within 10 ft, when being on non-overlapping and adjacent channels. This reduces data rates that can be achieved when sending to the APs in question, as they will have a lower SNR when receiving, thus causing clients to switch to lower order modulation rates, which will consume more airtime. Even if the decrease in SNR is only 3 dB, that is doubling the amount of noise energy, and is enough to potentially cause a downshift from any one modulation method to another. According to the relative constellation error chart in the 802.11-2016 standard in section 21.3.17.4.3 , depending on the FEC coding being used with the modulation. For example for QPSK 1/2, the required SNR is 10 dB, and for QPSK 3/4, the required SNR is 13 dB. If you had an SNR of 12 and then added another 3 dB of noise, you would have an SNR of 9, which would shift you down most likely. Now this does depend on your receiver, some radio chains are more sensitive than others, and can demodulate a much weaker signal, at varying SNR, at the varying data rates. You will find that many APs will have their receive sensitivity / receiver minimum input levels clearly listed in their datasheets / whitepapers. You will most likely not find what SNR goes along with that sensitivity, as that could vary from vendor to vendor, they only need to meet the requirements set in 21.3.17.4.3.


Time for logarithmic math as promised. If we have two APs 5 ft apart from one another, one on channel 1, the other on channel 6. And our TX power is at 20 dBm, with a 3 dBm omni antenna, our main lobe would have an EIRP of 23 dBm. Our first side lobe would be at 3 dBm EIRP, our second side lobe would be at -5 dBm, and our third side lobe would be at -17 dBm. Now lets calculate FSPL for these lobes.


FSPL= 20LOG10(1.52 Meters)+20LOG_10(2400000000)+20LOG10(4*π/300000000)= 43.68


We will lose 43.68 dBm to FSPL, if we subtract our antenna gain of 3 dBm for each AP, we make this 37.68 FSPL. So for our main lobe, the received signal would be approximately -14.68 dBm. The first side lobe would be received at -34.68 dBm, our second side lobe would be received at -42.68 dBm, and our third side lobe would be received at -54.68. This means that it would be possible that ACI would have the same exact affect as CCI, as -54.68 dBm is strong enough for CCA ED to detect and mark the medium busy. If we calculate the same thing but at ten ft.


FSPL= 20LOG10(3.048 Meters)+20LOG10(2400000000)+20LOG10(4*π/300000000)=49.72


Our third side lobe would now be at -60.68 dBm, still high enough to cause CCA ED to trigger a busy medium. So you would need approximately 4 meters / 13.1 ft between the APs assuming their was a clear line of sight between them, and no multipath was reducing signal strength. But if you are designing your WLANs for 5 GHz, then you are in better luck. Remember you lose approximately 6 dBm because of antenna aperture when we compare 2.4 and 5 GHz. Which means 10 ft or 3.048 m would work, in preventing this ACI from triggering CCA ED, but there would still be a potential noise floor increase.


Keep in mind that FSPL doesn't take any other conditions that would attenuate the signal into account, such as obstructions, or any of the RF behaviors like reflections, refraction, or diffractions, multipath. Or conditions that could cause a stronger signal, such as multipath, or TxBF, or antenna diversity with STBC or CSD and MRC. Nor does it take into account other characteristics of the radio chain being used, such as radiation patterns. Or perhaps a high quality radio chain could attenuate the side lobes even more. If there are any RF engineers that read this, I would love to pick your brain on a few things.


To put it bluntly, RF math like FSPL, is great on paper. But isn't perferctly accurate, you should think of it as an approximation at best. Which is why you should be measuring your wall DB loss for predictive surveys, and why you should always be doing a validation phase of a deployment.



Now lets take a look at some real life examples. For all examples, the AP remained in the exact same place. Figures 3 and 4 were generated by running a speed test with a Google Pixel 3 XL at approximately 20 feet away from my spectrum analyzer. And all distances were measured with a laser distance measure. First lets look at a baseline in Figure 2, this was taken over the course of several minutes, and at 5.61 ft from the AP, you will notice that there aren't any areas of the spectrum being heavily utilized, as red is >= 50% utilization, yellow is 30-50%, green is 20-30%, and blue is < 10%.



Figure 2

The first example we will look at is Figure 3 below. This figure represents my home WLAN, which is operating on channel 6 at 20 MHz. And I'm standing approximately 10 ft away from my AP and 20 ft from the generating device, in LOS of the AP and generating device. You can see a decently strong side lobe signal, just outside of channel 4 and channel 8, which wasn't present until I generated traffic. And a very small amount of side lobe energy outside of this. Take note of the noise floor in channels 1-4 now compared to our baseline.


Figure 3

Now lets take a look at Figure 4. This spectrum analysis was done at 5.61 feet. And you can see that considerably more ACI was generated in the side lobes. Extending all the way to channel 1 and channel 11. Adding to the noise floor for those channels.


Figure 4

The effect is even worse for wider channels. Please note that for the next measurements I made them considerably closer to compensate for antenna aperture. I would have done all of this in the 2.4 GHz range, but there are too many 20 MHz channels around, so my WLAN kept being forced back to 20 MHz. Figure 5 shows a 80 MHz channel at 4 ft away from both the AP and generating device, you can see there is considerable side lobe energy being generated.



Figure 5

In Figure 6 we see the same channel and traffic being generated but at 10 ft away from the AP, and 10 ft away from the generating device. You can see the side lobes are basically gone. The issue is clearly not as bad in the 5 GHz band, but the 2.4 GHz band suffers from this much more. I would attribute this mostly to antenna aperture.



Figure 6


As you can see from the examples though, FSPL isn't exactly accurate is it? We can see that through the practical examples from real life, we are no where near above the -62 dBm CCA ED threshold with most of the RF energy being generated. That doesn't mean at least some of the energy being seen by the AP isn't above the -62 CCA ED threshold, and causing the AP to consider the medium busy. But the real issue is that ACI causing an increased noise floor.


I decided to try one last example from further away with some light obstructions in the way. In Figure 7 my spectrum analyzer was 26 ft away from the the AP and the Google Pixel generating traffic. You can see that there is still plenty of energy in the side lobes, even at this range. It would appear this is part of why the noise floor tends to be closer to -90 dBm in most enterprise environments, instead of -101 dBm as thermal noise would dictate.



Figure 7


In conclusion, place your APs as close as needed, until your traffic is affected negatively by the increased noise floor (assuming you are using adjacent non-overlapping channels). And make sure to design your WLAN around 5 GHz. And simply turn off radios in the 2.4 GHz band where needed.











117 views0 comments

Recent Posts

See All

Comments


bottom of page