Fixing Hallway-Fi™, Vol.1: Single Story Building

Fixing Hallway-Fi™, Vol.1: Single Story Building

The school networks I look after have a common problem: the access points are all in the hallway, leading to massive co-channel interference (CCI). The only way to really fix this is a complete redesign, so my title might be a little misleading. For now, budget and time restraints dictate that I must make them work as well as possible as-is. Have lemons, make lemonade, right? Add some gin and call it Party-Fi.

So, for my first trick, I shall make an incredible amount of co-channel interference disappear. With the help of my lovely assistant Ekahau Site Survey (Vanna White) and the not-so-lovely master of ceremonies Juniper RingMaster (think James Franco hosting the Oscars), I’ll demonstrate just what putting AP’s in your hallway does to your network. Here is what your coverage might look like.

wes first survey initial deployment ss

That is a lot of green right? And green is good! Green means go! Go dog go! Or is it? Lets have a look at another important visualization: channel overlap. You can read about channel overlap in this excellent blog post by Andrew Campbell (@WiFiAndrew) of Ekahau.

wes first survey initial deployment co

Not so green anymore, is it? Not quite 50 shades, but I would call that about 95% grey. You may be asking, what does the grey mean? Well, the grey indicates an area in which a station can hear 3 or more access points on the same channel at a signal strength of -80 dBm or greater; there are at least three APs trying to share the same airtime on the same channel, as well as their associated clients. Some areas had 5 or 6 radios overlapping. The problem with hallway-fi is that there is no attenuation between the access points, so RF energy travels a long way, and co-channel interference is very hard to avoid. Here is an example from my initial survey, showing how the RF energy from the selected AP in orange travels down those hallways and reaches 9 other APs. If any of those 9 APs are on the same channel as the selected one, they are sharing airtime and may as well be one access point. Besides, how many of your clients sit in the hallways?

wes hallway example

If you haven’t counted the access points yet, there are 18, some of which are mounted vertically. The vertical APs really help keep the signal out of the classrooms as well, because of the antenna characteristics. I hope to post on that soon as well. Eighteen access points works out to just less than 1 AP per 2 rooms; there are about 30 rooms (shops, offices, classrooms) that house clients. It’s not quite the salesman’s bullseye of 1 per room, but it is still too many. How many should there be? I’ll get to that.

Another issue is the transmit power. All 2.4 GHz radios were set to 18 dBm (63 milliwatts). The 5 GHz radios were set to 15 dBm (31.6 mW). Both of those levels are too high, in fact they are the maximum for that model of access point (Juniper WLA532).

wla532 tx power

Most of my clients aren’t going to transmit at that power level at a desirable date rate (ie MCS5 or higher for 20 MHz/1SS), and the resulting cell sizes are too big for the number of access points. Those power levels will also lead clients to connect to the 2.4 GHz radio, as its signal will be stronger. Since it’s summer and no students are in school, a graph of client connections isn’t helpful to compare results.

One last thing: all legacy (802.11a/b/g) data rates were enabled. Those beacons at 1 Mbps are going an awfully long way. My LG G3 would not roam until it was across the building from where it first associated (although I believe Android clients are notoriously sticky); it just rate shifted lower and lower as I moved away from the associated AP. This is what the RF neighbor table looks like from one AP.

wes initial neighbours WES1

That access point could hear EVERY OTHER AP IN THE BUILDING! Some of you might remember this image from Eddie Forero’s (@HeyEddie) webinar with Ekahau entitled “The 7 Tricks of WiFi Professionals”.

So, in this school, we had too much co-channel interference for four reasons:

  • too many access points,
  • access points in hallways,
  • 802.11b and lower 802.11a/g data rates enabled, and
  • transmit power too high.

Depending on the network, some CCI in some places in acceptable. Andrew von Nagy (@revolutionwifi) presented at the last WLPC and talked about not being able to avoid some CCI in designs. CCI can be a pretty tough itch to scratch in certain deployments such as very high density, but I haven’t had that pleasure yet.

I started off with a survey to see exactly what was going on, the results of which are in the two heat maps above. Side note: I quit using continuous surveys in classrooms as there is just too much stuff to walk around. I still use continuous for hallways, but stop-and-go for classrooms. Attenuation values didn’t matter to me at this point, so I took four measurements in each room, around where students would sit.

My first configuration change, regardless of survey results, was to eliminate legacy rates. We have a few 802.11g clients left, but certainly nothing 802.11b. I also set the minimum basic (and therefore the beacon) rate to 24 Mbps for 802.11g and 802.11a. After taking a good look at the coverage of each access point, I ended up disabling half of them. Transmit power levels were set to 4 dBm on 2.4 GHz radios, and 12 dBm on 5 GHz. I have to manually steer clients to 5 GHz because Juniper’s band steering is part of their load balancing feature. I’ve found things work better with that disabled, and my sites don’t have a high enough density of clients to require it. After those changes, I went back out and resurveyed. Here is the signal strength map with 9 access points.

wes second survey post change ss

You can see there is not nearly as much green, a little yellow, and even some (GASP!) orange. Above is the 5 GHz coverage. The 2.4 GHz coverage is a little less good, with grey in the bottom left corner and middle portion just above the AP on channels 1 and 157. I plan to address the grey space in the center (gymnasium) with an AP and directional antenna. I ended up making a change in the upper right hallway, which I’ll go over shortly. Next is the 2.4 GHz channel overlap of the new configuration.

wes second survey post change co

Lots of green on that channel overlap map, which in this case is a good thing. The channels that do overlap are not the same as the strongest AP in the area, so the effects should be minimal. Here is the 5 GHz overlap.

wes second survey post change co 5

Those are 40 MHz channels in 5 GHz. So for the changes in the upper right hallway, I disabled the two APs that were on and enabled the other two that were in that hallway. I kept the channel plan the same and moved them, for lack of a better phrase, to the left. The rooms at the top end of the hall on both sides are computer labs full of desktops, so some poor coverage there is not a problem. Here are the survey results from after those changes.

wes mid years re-survey

I only resurveyed the hallway in question and a bit beyond, and I was happy with the results. The new channel overlap was basically the same.

Now, did I have enough access points? I pulled up the Capacity Planner from Revolution Wi-Fi to find out. This is a great tool. If you haven’t checked it out, I highly recommend you do so, as well as the blog. It really takes the guessing out of the capacity portion of a WLAN design. Here are my input and results.

wes capacity planner inputs wes capacity planner results

I know, I know… Wire up those Apple TVs; working on it. Either way, you can see with 9 access points I should be doing OK until we upgrade and redesign.

The final thing I wanted to check was how an iPad would connect in the rooms that had signal levels in the -70’s (the orange areas, specifically the bottom left of the map). I grabbed an iPad 2 running iOS 8.3, found this post about wi-fi scanning in iOS by George Stefanick (@MY80211), and downloaded the nPerf app for testing. Here is the setup in the room. Pretty simple.

test setup

The iPad scanning results looked like this.

WES compare resource ipad

The Ekahau signal levels looked like this. I’ll call those close enough to make no difference. A former coworker of mine used to say, “What’s a couple dB between friends?”.

wes compare resource ekahau

The nPerf results on the iPad were OK. In this particular room, there will only ever be 3 or 4 iPads connected using core apps; no Apple TVs or 802.11g laptops.

WES compare resource ipad nperf

For comparison, here is a test from directly under an AP with  a received signal level of about -40 dBm.

WES compare mid years hall ipad nperf

Granted, tests to the internet, and throughput tests in general, may not be the best indicator of network performance, especially with the WAN we have.

So, to optimize this hallway “design” and minimize co-channel interference, I did the following:

  • turn off legacy rates
  • turn off half of the access points
  • turn 2.4 GHz transmit power WAY down
  • slightly reduce 5 GHz transmit power
  • turn off Juniper’s auto-channel setting and do a manual channel plan

All in all, lots of turning things off. It’s a shame when you consider doing a proper design in the first place would have showed only 8 dual-band APs were needed, instead of 18! I am pretty happy with the results of the changes I made, and we’ll see how things work when September rolls around and kids are back in school.

For more reading on dealing with hallway designs, check out Jake Snyder’s (@jsnyder81) blog for 3 posts on the subject.

Thanks for reading. Please comment if you have anything to add or share, or have any questions!

  • Steve,

    Thanks for sharing your experience with this school.

    I’ll follow along your same outline as the post and make a couple of comments along the way that might help in future engagements.

    Co-Channel Interference – when on the ESS ‘channel overlap’ visualization, don’t use the default of -80dBm, try pushing it down to -85dBm. Next, if you leave the default color palette, you get the ‘shades of grey’ you referred to. Instead, use the adjustable color palette, change the limit to 6-7, and then set the color change handle so you get a color gradient, green for 1 AP on the frequency, yellow for 2, orange for 3, and red for anything above 4 AP’s sharing the same frequency.

    This way – your CCI graphic would pop a bit more and it makes it easier for those reviewing your heat maps to understand the severity of CCI a bit more than the subtle grey changes of the default palette.

    It would be helpful to know what the color palette was for the graphic showing the single AP coverage – what was your ‘want’, ‘don’t want’ and ‘don’t care’ set to?

    Want – the target RSSI and the ‘edge’ of your color palette (example -67dBm)
    Don’t Want – the color grey in the palette – greater than your target, but less than the CCI threshold. (-67dBm to -85dBm)
    Don’t Care – the area where ESS shows White – (example <-85dBm)

    Wow… all these AP's were mounted vertically! That in itself if an entire other blog post… they really did a job on this site eh? Both hallways AND vertically mounting AP's incorrectly. I've found many installers just do the vertical mounting because they see the 'keyhole' mounts on the bottom of the AP's and incorrectly assume that is what they are there for. Sad… really sad.

    Normally, you'd set the Tx power on the 5GHz bands to be at least 6dB stronger than the 2.4GHz in order to have nearly the same cell sizes between frequencies. (Of course you have to stay within proscribed legal limits based on regulatory domain and band)

    In this size school, you should be able to maintain near maximum Tx power on the 5GHz channels without causing any undue CCI – by using 20MHz channels and using as many unique frequencies as possible. (assuming the US regulatory domain)

    Turning off legacy data rates is a good thing. A very good thing. But it does NOT reduce cell size at all for CCI purposes. The Preamble Detect effect of Clear Channel Assessment on nearby AP's and Clients will still 'see' the AP's at a distance, just not be able to send data at those distances. So the CCI overlapping zones is just as large as it ever was – turning off lower data rates does not reduce that cell size in the least.

    What it does, however, is to help push clients to roam to a better AP choice sooner and addresses some of the stickiness you referred to.

    The biggest reason you had, and still have, CCI in this school is the AP placement in the hallways. I didn't remember reading anything about how you addressed the hallway placement. Were you able to move the AP's out of hallways and into the classrooms to improve the CCI overlapping zones?

    When adjusting Tx power on CCI-inducing AP's… I've found it a much superior solution to turn OFF an AP's 2.4GHz radio rather than merely turning the power down. Turning the power down still causes much the same CCI – you can test this thesis in ESS by showing how the CCI "Don't Want" – grey zone doesn't change all that much with Tx power adjustments. To get rid of most CCI you'll need to turn OFF the radios, not merely turn them down. It is all about Preamble Detect in the CCA – not whether or not a client can associate and send data at distance.

    Setting minimum data rates is more of a 'tuning' feature… best used with live client data. I've found setting minimum data rates too high can also cause a reduction in net aggregate throughput. Some sites, with some client loads can not maintain high net throughput by merely setting minimum data rate to say, 24Mbps… run a test to see where the 'sweet spot' is for your particular site. Some may be 12Mbps, some 24, and other will run best at 36Mbps… it pays to test.

    As you so aptly mentioned, CCI is the killer of your Wi-Fi network throughput – so running Andrew Non Nagy's Capacity Planner is a great start… but it matters not at all what the capacity planner says you need in AP counts if those AP's don't have frequency re-use. If two or more AP's share the same frequency… they share the frequency's available bandwidth. So without frequency reuse, you do NOT get the additional capacity of that next AP. If you have CCI between two AP's – you have the net capacity of only ONE AP!

    As you noted, I too have found the Juniper auto-channel to be nearly useless – so using manual channel plans was a great suggestion.

    For your optimization – I would have done the following:
    – Turn off legacy data rates
    – Turn off 1/2 – 2/3 of the 2.4GHz radios (leaving the 5GHz radios on)
    – Set the 5GHz radios to use 20MHz channels to minimize possible 5GHz CCI
    – Left the remaining 2.4GHz radios on high power to keep coverage up
    – Leave the 5Ghz Tx Power set to maximum
    – Manual channel plan for the 18 5GHz AP radios, and perhaps leave on 6 of the 2.4GHz AP's
    – Move all the AP's out of the hallways and into the classrooms wherever possible

    Thanks for a well-thought-out and supported blog post!



      Keith, thanks for your comments! I’ll answer as best as I can.

      The color palette for the CCI map was the default – I will take your advice and begin using the “want/don’t want/don’t care” method. It makes a lot of sense.

      The color palette for the single AP coverage was:
      -Want: >= -72 dBm
      -Don’t Want: -72 dBm to -80 dBm
      -Don’t Care: -80 dBm
      I can see that I should change the “don’t want” and “don’t care” levels a bit to better reflect preamble detection. I should care about that -80 to -85 range.

      Not ALL the APs were mounted vertically, but enough were. I am planning a post on that topic! This school is one of 14 in the division I work for. The story is that my IT colleagues, who are great at lots of other stuff, had the typical “more WAPs=more coverage no matter what” idea about wireless. They are learning and I am handling all the wireless now anyways.

      I haven’t been able to get APs out of hallways yet. The construction of this school makes that a larger project than we have time and money for right now, so I’ll do a redesign when we refresh and change vendors.

      I did turn off half of the 2.4 GHz radios, but the others really were still needed for adequate coverage for older single band laptops we still have in service.

      Thanks again Keith, for the very helpful comment!

  • Good job Steve!

    Great comments and replies! Love to see people working together constructively.


    Darby Weaver