Monday, 12 May 2014

Performing A WiFi Survey For A Building That Doesn't Exist (Yet)

I was recently involved in some work for an organization that had a newly constructed building to expand their, already sizable, campus. The building was a new, state-of-the-art facility that would require ubiquitous WiFi network coverage for a range of devices, including wireless voice handsets.

Like many organisations, they were faced with the challenge of specifying the wireless equipment that they would require long before the building had even started construction. In order to secure the funding they would need for the new wireless infrastructure components (i.e. APs, wireless controllers, licensing), they had to try to anticipate what the new wireless network would look like, taking account of the services that they would require.

The organisation realised that they would require a few more access points than might be usually expected for basic data services over wireless, but had taken a best guess at how many APs they might require and where they might go. When we walked in to the new facility with all of the APs deployed as per their "best guess", it was immediately obvious that the APs were all too close together and there was massive co-channel interference throughout the facility. Turning off a few radios relieved some of the immediate pain, but it was obvious a major re-think of the 'design' would be required, with many APs requiring a move or even removing altogether.

In a brand new facility that had just been handed over from the building contractor, this was definitely not good news. It's amazing how unpopular you can become when you have to move APs and juggle brand new ceiling tiles that already have holes drilled through them to for AP data cables. The same applies for breaching and making-good fire-breaks between rooms in a new-build, just because you need to move an AP between rooms to a new position.

This experience reminded me that not everyone may have heard about predictive wireless surveying, or even if they have, they may not understand its value or how it is achieved.

In summary, a predictive wireless survey allows complete a wireless survey of a building or area by using scaled plans of the area that requires WiFi coverage. It requires specialist survey software that allows building plans to be imported, obstructions (such as walls) to be added to the plans and 'virtual' APs to be added to the plan. The survey software then performs some very fancy calculations to give a modelled view of what the WiFi coverage of the APs might look like.

This is obviously a great approach when having to design a wireless network for a building that hasn't even been built yet. By carefully modelling the environment that the wireless network will be deployed into, a reasonably accurate network can be designed just from building plans. This method isn't going to create a survey report with the same accuracy as a physical survey performed on-site, but it's going to give you a very good indication of the final network is likely to look like. It will certainly allow you to build a very accurate bill of materials for the final deployment.

Although there will generally be a charge for a survey of this type, the cost is likely to be very minor compared to the costs of over or under-provisioning that a 'best guess' approach will yield. There are also the additional costs down the line around re-deployment, additional cabling, additional licensing etc. that you will avoid with a well planned design method. The cost of a predictive survey is likely to be dwarfed by the cost of an poorly designed/anticipated wireless network - this could be many thousands, or tens of thousands of dollars in wasted hardware, licensing and man-hours, compared to the very modest cost of a predictive survey.

The one caveat I would advise when choosing a supplier to provide a you with a predictive survey (also known as a desktop survey) is to use one that uses staff with at least CWNA certifications. Many less-qualified suppliers may get hold of a copy of wireless survey software and 'give it their best shot'. But, to use the tools effectively, you really do need to understand RF propagation, WiFi design principles (such as capacity planning) and best-practice approaches (e.g. avoiding co-channel interference, hidden node issues etc.). Without wanting to labour the point too much, don't be swayed too much by suppliers who claim to have 'vendor-level' certifications that enable them to perform survey work - the vast majority still have very little in terms of survey knowledge and a pitiful level of RF ( have be warned).

A Predictive Survey

So, how do we actually perform a wireless survey for a building that doesn't exist yet? I'll present a number of screen-shots from the Ekahau wireless planner survey tool. The Ekahau survey tool allows both on-site surveys using a traditional "AP-on-a-stick" surveys, together with (predictive) desktop surveys that we have discussed above.

To perform a predictive survey, one thing we absolutely must have is building plans of the area to be covered by the new wireless deployment. Without accurate building plans that can be scaled to represent the final real-world building, it is not possible to accurately model the final environment and correctly show the coverage and performance of the WiFi network.

Once we have an electronic copy of the floor-plans for our survey area, we can import it in to the Ekahau tool. The first thing was have to do is to calibrate the plans so that we have an accurate indication of the size of the area being surveyed - this is usually done by entering a measurement for one wall of the building. Without this initial accurate measurement, the predicted coverage areas for the survey are pretty much meaningless. The software calculations for RF signal levels rely on the distance of each area on the floor plan from the access point - if the distance is incorrectly calibrated then the RF coverage will be inaccurate.

Fig 1 - Imported Floor Plan (with calibration of one wall shown)

In the example we're going to look at, we're going to do a very quick and dirty survey to show the principles of the predictive technique. In our case, we're going to be looking for coverage in all office areas, providing basic RF coverage down to -67dBm in all areas on the 5GHz band. We're looking for an SNR of 25dB (or better) in all areas too. A real survey would be a little more specific than this, but this will be good enough for the purposes of our example.

The next thing we need to do is to draw in the building structure of the floor plan to build up a model which will start to reflect the actual construction of the building (as closely as we can). Walls, doors, windows, lift shafts etc. can be added across the floor plan by tracing their outline with the appropriate building material that is provided in the survey software. The survey software provides typical losses that will be experienced through each type of material so that the RF losses around the building may be calculated by the survey software.

In our example, I've added in the outline of the building, together with some internal walls and doors (the gray and brown lines overlaid on the plan).

Fig 2 - Floor plan with building materials added

Once we have added in the building structure, our next step is to start to add in some access points to start to see the coverage we might achieve as we place them in various positions. The first thing that I need to do is to select the AP and antenna that I intend to deploy. The Ekahau tool has a massive range of access points and antennas to choose from. In this case, I'm going to select a Cisco AP2600i, which has an internal antenna.

In the screen-shot below, I've placed an AP in the top left of the plan to see the coverage that it gives me. The graduated colours show the signal strength as the RF signal radiates out from the AP, going from high (green) to low (red). Where the coloured coverage pattern ends is where edge of our cell may be found - note the legend in the bottom right of the graphic which shows our -67dBm cut-off selection (which is one of our design criteria):

Fig 3 - Initial Access point placement

This coverage is probably a little more than I would ideally like, so I'd like to reduce the coverage area of the AP. I can quickly edit the transmit power of the AP and see the effect on coverage as I wind the power down by 3dB:

Fig 4 - Initial Access Point Placement (reduced power)

I can now proceed with adding a few more access points to give me the level of coverage I'm looking for. As well as adjusting the transmit power for each access point, I can also assign channel number for both the 2.4GHz and 5GHz bands. This is useful in assessing my co-channel interference, to understand if I can achieve a viable channel plan across the building.

In the screen-shot below, you can see a number of APs placed across the floor-plan with the appropriate channels settings displayed for both bands:

Fig 5 - All APs placed showing signal coverage for 5GHz

Now that we've placed our APs, we can see the coverage that they would provide when deployed in these positions. We can now look at other aspects of the predicted deployment to see if it meets our design criteria. For instance, do we meet the 25dB SNR requirement across the floor? If we switch the display filter to show signal to noise ratio (SNR), we can see what the predicted SNR might look:

Fig 6 - Signal to Noise Ratio

Again, looking at the legend at the bottom right of the screen-shot, we can see that our lower cut-off (which is a design minimum) is set to 25dB.  The red areas show where we are close to our lower limit. Although we are close to our limit in a number of areas, we nonetheless meet the design criteria in all office areas.

Finally, we'll do a quick check to see if our channel plan is causing any co-channel interference between access points . Co-channel interference (CCI) is caused when a coverage area has two or more APs on the same channel that may be heard by a client in that area. CCI is undesirable as it causes contention across the APs and clients within that area, lowering the overall throughput on that particular channel. In our example, we can see that we have no co-channel interference on the 5GHz band (again, note the legend showing that only 1 AP can be heard in all areas):

Fig 7 - Co-channel Interference on 5GHz

As this doesn't make a very interesting result, here's the same screen for the 2.4GHz band. The yellow blobs indicate where we might experience co-channel interference. We might like to look at moving our APs, adjusting power levels or maybe even turning off some of our 2.4GHz radios if we decide that this an issue for us:

Fig 8- Co-channel Interference on 2.4GHz

Well that's pretty much all I want to cover around doing a predictive survey. We've really only scratched the surface of what is possible with predictive surveys, but you can hopefully begin to understand how we can build a model of the RF environment for a building and then how a WiFi network would be deployed and might perform in that environment. There are many more aspects that we could consider around other factors such as capacity planning and 3 dimensional building planning (for multi-floor deployments), but we'll leave those for another time (or, even better see further examples over on the Ekahau web site)

It has to be emphasized that the results of a predictive survey are only as good as the data that is supplied to the RF model that is built. Unless the walls and obstructions traced on the report are accurate, then the clever RF calculations performed by the survey software  to show RF coverage and other characteristics will be inaccurate. 

A predictive survey is generally unlikely to be as accurate as a physical survey conducted on site, but it certainly provides a very good indication of what might be expected for a building that has yet to be built. This can be invaluable for putting together an initial bill of materials, together with estimates for cabling and switch infrastructure requirements for the new wireless network.

Hopefully, this has given you a 'taster' of the value of predictive surveys for buildings that have yet to be built, or perhaps buildings that are undergoing a major refurb and are not accessible at the time when new kit needs to be specified and ordered.