As society has moved towards ever smaller and more mobile devices, Wi-Fi has become an essential connectivity standard which we rely upon in our daily lives. It was initially made popular in the home by the availability of affordable, standardised 2.4Ghz equipment under the 802.11b/g/n standards. The 2.4Ghz band is far from ideal with limited bandwidth and high interference potential. Demands for higher bandwidths for higher throughput, improved reliability and resilience against interference, paired with rapidly reducing costs of dual-band radios have pushed many users to using the 5Ghz band as well. The 5Ghz band was initially less popular owing to higher equipment costs and lower propagation, but offered many more channels to choose from and potentially higher rates in not having to deal with “legacy” clients.
With the advent of even wider-bandwidth Wi-Fi modes such as 802.11ac Wave 2 with greater affordability, it seems that the initially spacious 5Ghz band may soon become just as crowded as the 2.4Ghz band is today. As a result, it’s probably important to give channel selection a thought for optimal performance, especially because it can get quite confusing.
Radio Rules: The Letter of the Law
Radio signals are electromagnetic waves that propagate in open space. Radio signals of a given frequency, bandwidth and power can travel significant distances and cause interference to other signals. Because of the shared aspect of the radio spectrum, most countries regulate the use of radio spectrum through a government body. In the case of Australia, this is the Australian Communications and Media Authority (ACMA). Wi-Fi equipment sold in or operated in Australia operate under the Low Interference Potential Devices (LIPD) class license and thus does not need to be individually licensed provided they meet the rules in the legislation. At present, the legislation was last updated in 2015, and contains a section for “Frequency hopping, WiFi and RLAN transmitters” which details the rules under which devices can operate under the LIPD class license.
Much of the Wi-Fi frequency spectrum owes its existence to exceptions made in frequency planning across a number of countries for “industrial, scientific or medical” (ISM) usage. These frequency bands enjoy an exception because they usually involve processes which generate significant radio energy (e.g. microwave ovens) but do not require any protection against interference. As a result, adding a little radio energy to these bands for use with Wi-Fi wouldn’t adversely affect the operation of ISM devices in general, but would bring about utility in enabling the provision of Wi-Fi-based services (especially where the ISM band isn’t actively polluted or by coexistence measures). The 2400-2500Mhz band and 5725-5875Mhz band are both ISM bands.
However, additional spectrum in the 5Ghz band can be had by using other frequencies. In the US, they have an unlicensed national information infrastructure band (also known as U-NII) which was opened to Wi-Fi 802.11a systems under certain restrictions. This band is often used for radar systems – stringent power output limits are enforced by transmit power control (TPC) and stations are required to detect and avoid radar through dynamic frequency selection (DFS). Where one country makes certain exceptions, other countries often consider aligning their own rules over time (where possible) to reduce administrative burden and allow for easier use of imported equipment. This gives us further channels to use for 5Ghz Wi-Fi.
However, because of this issue, the channels, power levels and operating modes available in each country can be quite different. Unlike 2.4Ghz where the rules can be simplified mostly to “US, Japan and Rest of World”, 5Ghz Wi-Fi requires a lot more complex logic to ensure compliance. With Wi-Fi equipment often travelling overseas (through importation, immigration, business trips, holidays), ensuring compliance has resulted in a mixture of strategies. Devices can be configured for a regional domain directly, devices can have a coded regional domain in their EEPROMs from where they were sold and devices can receive regional domain information over the air from access points in their beacons through an embedded “Information Element” (IE). Because of the potential issues with misconfiguration, most regulatory domain ambiguities are resolved by choosing the most restrictive subset of channels that comply with all the above.
While most regulatory authorities dealing with spectrum prefer to refer to frequencies and frequency ranges, Wi-Fi equipment often uses a channel number for simplicity. In the case of the 5Ghz band, the center frequency of a given channel number is:
F (Mhz) = (5Mhz * channel number) + 5000Mhz
As 802.11a (the initial standard to use the 5Ghz band) used 20Mhz OFDM modulation with 16.25Mhz used by carriers, non-overlapping channels are separated by 20Mhz or 4 channels. In a bit of foresight, most equipment manufacturers avoided a repeat of the 2.4Ghz channel selection nightmare by limiting the channels the user could choose to those which did not overlap. This resulted in most equipment providing channels in steps of four – e.g. Ch 36, 40, 44, 48.
In the case of 802.11a, this made things simple – each channel that could be selected was independent of every other channel that could be selected. But the introduction of 802.11n changed this. The 802.11n standard introduced 40Mhz mode transmissions with 33.75Mhz used by subcarriers, thus consuming two contiguous selectable channels. One of the channels would be designated the “control” channel which would always be used, the other would be the “extension” channel which was used if the receiving equipment supported the bandwidth and if it was clear for transmission at the time.
The way this was presented as a selection differed depending on the equipment – Ch 36 + Upper, Ch 38 (40Mhz) and Ch 36-Ce (e.g. Mikrotik) would all be equivalent – denoting the use of Channel 36 as control and 40 as extension. Likewise Ch 48 + Lower, Ch 46 (40Mhz) and Ch 48-eC would all denote the use of Channel 48 as control and 44 as extension.
In this way, 802.11n allowed for correct operation if each channel was occupied, but if one access point (AP) were to use Ch 36 + Upper and another nearby were to use Ch 40 + Lower, there is a good chance that throughput will be impacted as neither could use the extension channel when the other AP was using it. As a result, it’s best if you can avoid overlapping any control AND extension channels.
The first wave of 802.11ac did exactly the same thing again, broadening the transmission mode to 80Mhz. As a result, one control and three extension channels are nominated in a contiguous block. Many routers still offer selection by 802.11a-style channel numbers and determine the appropriate extension channels automatically where possible. So in the 802.11ac case of Ch 36, this would mean using Ch 36 as control and Ch 40, 44 and 48 as extension. This could be denoted as Ch 42 (80Mhz), or Ch 36-Ceee.
Finally, 802.11ac Wave 2 does this yet again, but with the understanding that 160Mhz mode may not be possible in some regions due to a lack of contiguous spectrum. As a result, 802.11ac Wave 2 features a split 80+80Mhz mode as well. I haven’t experienced any such gear myself, but I suspect the channel selection logic is similar again – so Ch 36 in 160Mhz mode would occupy Ch 36 as control and Ch 40, 44, 48, 52, 56, 60 and 64 as extension and could be referred to as Ch 50 (160Mhz) or Ch 36-Ceeeeeee. Split mode would probably allocate channels in a near-identical way to the initial 802.11ac equipment.
Choosing a Channel
As mentioned in the previous section, the way channels are denoted can be quite confusing. For simplicity, I will stick with only 20Mhz wide channel numbers (802.11a) and use the Mikrotik style notation for control (C) and extension channels (e) where necessary (e.g. Ce, eC, Ceee, eCee, eeCe, eeeC, etc).
In Australia, the LIPD often defines the transmission power as a given mW EIRP per Mhz bandwidth and 20Mhz is often the basis of the calculation due to the width of the channel (especially when operating in backwards compatible mode where extension channels may not be available). This means equivalently, the power limit can be stated as 200mW EIRP in the first block, 1000mW EIRP in the second and third blocks. Note that EIRP stands for effective isotropic radiated power, so on a 2.1dB “rubber duck” omnidirectional dipole, your input power should be no more than 123mW or 616mW respectively, or less if you have an antenna with any gain. The final block is listed as a hard EIRP limit with no bandwidth dependency. The number in brackets indicates the type number in the LIPD legislation document where the limit is written.
Consideration 0: What are you allowed to do?
This broadly just means complying with the law and what the regulatory authority allows you to use. Pick a channel in the table above, assuming your Wi-Fi gear has the correct country code selected and regional domain information programmed in, and you should comply with the rules. This is important to avoid the possibility of a fine or degrading service for your neighbours, as the unlicensed spectrum is “shared” for the benefit of all.
Consideration 1: What are your neighbours doing?
Something that should never be overlooked is the need to do a site survey. This would entail using either your router (if it supports it), a smartphone or a laptop which supports 5Ghz band with a good antenna on a walk-around with Wi-Fi scanning software that can tell you the channel(s) used by your neighbouring access points and their strengths. This could be something like Vistumbler on a PC or WiFi Analyser on Android, noting that many apps will give you limited information especially on channel widths.
Mark down all of the data you can get to have an idea of which channels are in use. If you’re in a particularly crowded area, it may pay also to use more sophisticated tools (e.g. a wireless card in promiscuous mode, Mikrotik device in Snooper mode, etc.) to ascertain what level of loading each channel is experiencing in the case you’re expecting to run co-channel with another AP, so as to choose the one with least use/noise.
There is a good chance that in Australia, you will find that Ch 36, 40, 44, 48 and 149, 153, 157, 161, 165 to be somewhat busy. These channels were quite popular amongst older gear as they come with few to no catches (see next point) and existed prior to the 2nd and 3rd blocks being approved for use in Australia. Instead, consider choosing a channel where none of your neighbours are operating.
Consideration 2: What are you doing with your link?
Because certain channels have certain rules attached to them, depending on what you’re doing with your link, you may not be permitted to use certain channels or you may not want to use certain channels. This may be because of power limits or the need for Dynamic Frequency Selection (DFS) or Transmit Power Control (TPC).
The downside of DFS channels is that your wireless gear must comply with DFS operating rules. This means that it must listen for radar signals and change channels when it observes one. This means that upon starting up your AP, you won’t have a 5Ghz signal for a few minutes as it listens to determine whether to operate on the channel or not. Once it begins operation, it is still listening out for radar signals and can change if it hears one – this results in dropping all the clients and sitting on the new channel for a few minutes to decide if it can be used. This means that frequency coordination can be problematic and reliability of service can be poorer.
That being said, you could get a “heads up” by doing a search on ACMA’s Register of Radiocommunications Licenses. Selecting 320km radius from Sydney, searching the DFS channel frequency ranges of 5250 to 5730Mhz, I can only find one radar transmitter (Bureau of Meterology at Kurnell) operating at 5625Mhz with 1Mhz bandwidth. This is below Ch 132 and wouldn’t impact on Wi-Fi at all. The other transmitter is for uplinking to Pivotel satellites at Dubbo at 5170.5Mhz with a bandwidth of 159Mhz. Due to the radiation angle of this uplink, it really wouldn’t make a difference to us, nor would we make much of an impact on them. There is also the possibility of false radar detection due to misdetection due to noise generated in co-located gear or interference, so even if you’re not in a radar-coverage area, you may still suffer as a result of choosing a DFS frequency.
Power limits are also deserving of mention – the first block is limited to (effectively) 200mW EIRP, the second and third blocks are limited to (effectively) 1000mW EIRP and the final block is limited to 4000mW EIRP. As a result, if you had an omnidirectional antenna around 28dBm transmit power, you would not really be restricted in transmit power choosing the second and third blocks, but choosing the first block would require reducing your transmit power below maximum. If your equipment has an omnidirectional antenna and about 21dBm transmit power, you would not be restricted regardless of the chosen frequency. It is rare that higher transmit powers are particularly useful – but adding an antenna with gain reduces the permitted transmitter power accordingly, as the standard is specified as EIRP.
Case 1: Outdoor Long-Range Links
If your link is used for outdoor usage – say point-to-point (PtP) or point-to-multipoint (PtMP) networking, the final block (Ch 149-165) is most attractive because of the higher operating power available and no need for DFS thus ensuring fixed frequency plans and no possibility of unavailability for minutes due to start-up radar detection. Because of the “narrow” beams used in most PtP or PtMP links, frequency reuse may not be such an issue as the noise from co-channel users may be sufficiently attenuated. If you’re really in a pickle, you could use the 2nd or 3rd block frequencies, although with limits on power and possible DFS annoyances.
Case 2: Outdoor Wide-Area Coverage
If you’re dealing with outdoors, you’re not permitted to use the first block of frequencies. The final block (Ch 149-165) is, again, more attractive because of the higher power and non DFS requirement, but because they would be popular, it is probably more appropriate to use the 2nd or 3rd block where the gain and power level of your AP allows you to comply with the lower EIRP permitted, if you’re confident that no radars are nearby and the AP rarely loses power or is reconfigured.
Case 3: Mixed Indoor/Outdoor Coverage
Technically speaking, you’re not permitted to use the first block of frequencies for outdoor coverage (although, I suspect some home users walking outside with their phone wouldn’t know it). For mixed coverage, the final block (Ch 149-165) is more attractive because the higher power permitted may be necessary to penetrate building materials. If this is not required, then choosing the 2nd or 3rd blocks may be quite advantageous, as with the AP indoors, it is less likely to hear a radar or be subject to sufficient interference to cause false radar detections. Most average APs will not have a transmit power high enough to suffer any restrictions on the 2nd or 3rd blocks which have a 1000mW EIRP limit, thus Ch 100-116 and 132-144 seem quite good candidates.
Case 4: Indoor-Only Coverage
In the case of indoor-only coverage, all frequencies are available for use. DFS and TPC channels may be used and due to location, it seems rather less likely a radar hit would occur, thus reducing the likelihood of unscheduled frequency changes due to DFS.
However, depending on your equipment, choosing an indoor-only frequency may not be the best choice because the EIRP limits may cause your equipment to limit its transmit power below its maximum, reducing the coverage area or reducing SNR and throughput. The lower channels 36-48 are also quite popular amongst older equipment which cannot or does not implement DFS/TPC, thus may be noisy. The newer Ch 52-64, 100-116, 132-144 are much less often used and the latter two ranges can be used mixed indoor/outdoor.
Consideration 3: What mode are you operating?
Knowing the above, the mode that is used is important to determine how many clear channels we need for optimum performance and the limitations of each block of channels. The table above has number/letter coded “blocks” of the appropriate width for each mode, with no indication of which is control or extension.
For 802.11ac Wave 2 in 160Mhz mode, this requires eight contiguous 20Mhz channels, resulting in just one block that can be used – the one labelled 0 from Ch 36-64. This results in the requirement for indoor-only operation, DFS and TPC.
For 802.11ac in 80Mhz mode (or Wave 2 in 80+80Mhz mode using two blocks), this requires four contiguous 20Mhz channels which can be arranged in a number of ways from 1 through to A. The most efficient packing of blocks is represented by the numbers 1, 2, (3 or 7), 4, (5 or 8), thus representing five non-overlapping 80Mhz systems within the permitted channels. Of these, 1, (5 or 8) represent non-DFS selections, with 2, (3 or 7), 4 representing DFS/TPC choices. However, as 1 and 2 are indoor only choices. Selections of 6, 9 or A are suboptimal as they will cause a reduction in the number of simultaneous non-overlapping systems within the channel plan. These also bridge across non-DFS and DFS parts of the plan, thus will be subject to DFS rules regardless.
For 802.11n in 40Mhz mode, two contiguous 20Mhz channels are required, which can be represented by the letters B through S. Of these, the most efficient packing is from B, C, D, E, (F, G or O, P), H, I, (J, K or R, S). This results in the possibility of ten non-overlapping 40Mhz systems within the permitted channels. Of these, B, C, (J, K or R, S) are non DFS, with the remainder subject to DFS/TPC. Selections of L, M, N and Q are suboptimal as they reduce the number of non-overlapping systems available.
For 802.11a in 20Mhz, any channel from the list can be used independently for a total of 22 independent non-overlapping systems.
The above does not consider which channel from within a block is denoted the control channel and which is denoted as extension. For example, in 80Mhz 802.11ac, if we chose the block labelled 1, this could be set as Ceee (Ch 36), eCee (Ch 40), eeCe (Ch 44) or eeeC (Ch 48) despite occupying the same band of frequencies. Assuming an overlap is inevitable, it is probably most advantageous to overlap with one ac Ceee (Ch 36) and the neighbour as eeeC (Ch 48) as this will mean that the most busy channel is separated as far as possible from each other. It also means in backward compatibility 40Mhz transmissions, they can operate simultaneously non-overlapping, unlike if a Ceee system is next to a eCee system as if both systems are busy, a 40Mhz contiguous channel width encompassing the C channel would not be available.
Consideration 4: Do you need to support overseas devices?
This is an issue primarily faced by imported devices, especially ones with EEPROM region codes and very strict drivers. This may mean that a device is forced to only operate on channels which are permitted in common with your AP’s region, the region code in EEPROM and the region set in the driver. From Wikipedia’s List of WLAN channels, the channels which are permitted for use (even with special conditions) across all listed countries are summarised above. The channels are 36 through to 64, which are indoor channels in Australia. Choosing one of these channels will ensure that even devices with regulatory domain restrictions can see and connect with your network in the 5Ghz band.
In all, it may not be possible to win everything, given all the considerations at hand, so it’s about choosing which channel best meets your needs.
It Doesn’t (Quite) Add Up: Marketing Numbers
It seems that dissatisfaction with Wi-Fi often leads people to upgrading their equipment. A common complaint is that the numbers never quite add-up. This is for a number of reasons:
- APs are often sold with a “large number” on the box which indicates “combined” maximum wireless physical link layer throughput across all bands in the maximum bandwidth and number of streams streams supported by the unit.
- Wireless clients only can use one band at a time – so an AC1900 device which offers 600Mbit/s over 2.4Ghz and 1300Mbit/s over 5Ghz could offer at most 1300Mbit/s physical layer speed to a 5Ghz client.
- Wireless clients may not support the same modulation/bandwidth or number of spatial streams supported by the AP. A laptop with an 802.11n card with two streams connected to an AC1900 AP would only be able to talk at 300Mbit/s (40Mhz, 2 streams, 802.11n) rather than the 1300Mbit/s (80Mhz, 3 streams, 802.11ac) it could deliver. The same laptop with an 802.11ac card with two streams would only offer 867Mbit/s (80Mhz, 2 streams, 802.11ac) at best.
- Physical layer rates do not account for overheads – this includes pauses for checking the air before transmission, forward error correction, packet headers/preambles, etc. The actual best-case throughput is often 50-60% of the physical layer speed. In mixed-mode operation, there are some overheads due to backwards compatibility as well.
- Real-word signals are corrupted by noise, fading, multipath, changing conditions, etc. As a result, as signals get weaker, wireless cards will negotiate the rate to maintain the connection as best as possible, with physical rates falling down to legacy physical-layer rates of 6Mbit/s on 5Ghz and 1Mbit/s on 2.4Ghz.
- The radio medium operates in half-duplex, so the available air-time is shared between transmission and reception.
- The radio medium is a shared medium, so the available air-time is shared between all clients (and co-channel users). Slower clients with weaker signals eat up more air-time to transact the same amount of data as faster clients with stronger signals and reduce the “total” capacity of a given system.
- In the case of interference that cannot be corrected by forward error correction, retransmission of packets may be required, further consuming air time and reducing the effective throughput.
- Sometimes, the throughput can be limited by your equipment due to a lack of CPU time on a router, firmware issues, USB connectivity, etc.
Because of all these factors, even if an 802.11ac 1300Mbit/s connection looks faster on paper than a Gigabit Ethernet connection, it very likely isn’t, offering only about 500-850Mbit/s at the most.
Bad Air: Single-Radio Extenders
In many people’s haste to improve their signal level, they purchase a cheap single-radio “universal” Wi-Fi extender – the sort that often just plugs into a wall socket and “works with anything”. While they can have their place, they often are used as a bad band-aid solution that might make things worse.
In short, a single-radio extender uses just one radio on each band to extend your Wi-Fi signal. In wireless extender mode, they connect to your main AP via Wi-Fi and broadcasts another network from itself. Because they are using just one radio on each band, they are limited to broadcasting on the same channel as your main AP! This is bad news for a number of reasons:
- If the extender is placed far enough away from the main router, it would not be connected at the highest rate and would be restricted to lower data rates which consume more air time.
- Any data going through the extender will consume at least twice the air time – one lot of air-time to carry the data from the main AP to the extender and again on the same channel from the extender to the device. As this happens on the same channel, the extender is consuming channel time that the main AP could be using to serve someone else.
- Extenders can have mismatched abilities to the core AP – for example, a single stream extender on a dual/triple stream AP would have even more severe impacts on airtime utilisation and reduction of net total throughput.
- A universal extender generally operates with no expectation that the AP supports proper Wireless Distribution System (WDS) 4addr headers and has to resort to interesting tricks to ensure devices behind the extender can access the network correctly. Such tricks include running a second NAT (a second subnet) on a separate SSID, proxy ARPing for devices behind the device, etc. This may lead to a non-seamless experience (i.e. changing between extender and primary network results in a break of open sockets, new IP address rather than seamless roaming). It can also break certain peer-to-peer applications that require open ports for NAT traversal.
- Additional network fragility where the extender itself is having issues, troubleshooting the network can become difficult. Applications which expect the whole network to be in the same subnet with the same SSID would not necessarily work (e.g. home automation stuff). In some cases, the extender can be having a weak signal, high packet loss, issues connecting to Wi-Fi from the core AP, but still seemingly offers a strong signal strength.
- Network security risks, in the case the firmware of the extender is vulnerable, it could expose the rest of your network to attack.
In general, most of the extenders are used in the wireless repeating mode which suffers from the disadvantages noted above. The best approach is always to use Ethernet cable to wire up additional access points – if they’re set-up as APs with the same encryption parameters, wireless cards can automatically roam to whichever is strongest. Having these APs on non-overlapping channels increases the throughput available.
A compromise to this is a dual-radio extender – this uses one radio to connect to the AP, and another to serve the clients on a different channel, thus avoiding the double-usage of airtime that a single-radio extender would have. Another thing is to site the extender where coverage is still quite good to ensure quality of service to the extender itself (and anyone connected to it). Extending your network too far can hamper neighbour’s networks and open yourself to security risks and interference. It still have issues with roaming seamlessly, as either NAT or Proxy ARP will be necessary.
As far as I’m concerned, these single-radio extenders are better used as APs on the end of an Ethernet backhaul with the channel set non-overlapping for seamless roaming or not at all.
Bandwidth Hog: Mesh APs
Some of those with deep pockets who may have been burnt by the single radio extender have instead gone the “whole hog” and gone for the new style “mesh AP”. This is essentially just like the dual radio extender I mentioned before, but designed for proper wireless bridging for a seamless connectivity network and a big price tag. Often sold as a pair or more frequently as a set of three, the idea is that these APs offer both seamless coverage and speed.
Such units have a base unit, featuring a few Ethernet ports, a router, a radio in each band for local client connection, as well as a second radio in the 5Ghz band that participates in a wireless backhaul link that the satellite units connect to. The satellite units are much the same in design, allowing for local client connection and passing the data over the wireless backhaul to the base unit. As the system is designed with proprietary extensions, each vendor has its own system, but it ensures that the whole network is one subnet and seamless roaming is possible.
Aside from the great expense, these mesh systems can be a problem because of bandwidth utilisation. Being particularly greedy, the sketch above shows what happens in the worst case of a mesh system with three nodes, using the most greedy channel selection logic I can think of.
The base station decides to take Ch 6 (20Mhz) on 2.4Ghz 802.11n and Ch 149-161 (80Mhz) on 5Ghz 802.11ac for clients, and uses Ch 36-64 (160Mhz) 802.11ac Wave 2 for backhaul, anticipating the two satellites could be completely loaded at some time. The first satellite decides to use Ch 1 (20Mhz) on 2.4Ghz 802.11n and Ch 100-116 (80Mhz) on 5Ghz 802.11ac for clients, with the second satellite using Ch 11 (20Mhz) on 2.4Ghz 802.11n and Ch 132-144 (80Mhz) on 5Ghz 802.11ac for clients. Just with this one system, I’ve used all 2.4Ghz non-overlapping channels and all but Ch 116 and 165 on the 5Ghz band!
This wouldn’t be good for neighbours who will have to operate co-channel with you and shows you just how much of a bandwidth hog a mesh based system could be if configured deliberately to maximise one’s own throughput.
Again, going with cabled Ethernet where available is probably going to give you better, more reliable performance while reducing cost (no need for “fancy” mesh APs) and bandwidth utilisation. Most of the time I’ve encountered mesh networks, I haven’t seen a clear reason why a mesh network would be necessary – most houses aren’t big enough to need them if a powerful and sensitive AP is used and sited in a good location (towards the center of the house, high up if possible, away from any large metal structures and bodies of water).
Wi-Fi is something we’ve come to depend on over the years and 5Ghz Wi-Fi is now almost commonplace. As a result of advances in 802.11 modulation, with wider bandwidth modes, the once capacious 5Ghz band is beginning to feel like the congested 2.4Ghz band. Additionally, different countries impose different rules on the use of 5Ghz spectrum, resulting in a complex web of power limits and frequency selection rules which I detail. Choosing the best channel can help improve your network performance and unlock the full capability of your devices, as can careful siting of your access point.
Cheap single-radio extenders are not a great solution and expensive mesh systems are needlessly expensive where you can run Ethernet as a backhaul between APs. In fact, I suppose many mesh networks are not entirely necessary in the first place and instead serve to consume even more of the 5Ghz spectrum, making co-channel operation an unfortunate side-effect.