Saturday, April 16, 2016

Channel Bonding in Wi-Fi

Note, this blog was written for and appeared as a three-part series in Network Computing in January and February, 2016:

Channel Bonding In WiFi: Radio Frequency Physics

(Appeared in Network Computing on 1/11/2016)

Using larger channels improves WLAN capacity, but the laws of RF physics can pose challenges when putting channel bonding into practice.
In the WiFi industry, we are on a never-ending quest always to push more and more data to more and more devices. A seemingly great mechanism for this ischannel bonding, where neighboring 20 MHz channels are bonded together to form a larger channel. By doubling the channel width, you get slightly more than double the data capacity of the transmission.
Alas, however, as with all throughput enhancement techniques, channel bonding comes with a price. Implementing channel bonding in ever-increasingly complex WiFi networks is a careful balance of tradeoffs that can make it difficult for even experienced WiFi professionals to properly understand. The natural inclination is to make the channels as large as possible, and most access point vendors make it very easy to do this on their equipment.
However, what generally is not provided is guidance as to where and how larger channels should, and should not, be used. As a result, many modern WiFi deployments are problematic, because their environments cannot properly handle the use of larger channels. 
In a series of blogs, I'll discuss the various tradeoffs involved in channel bonding, cover the regulatory restrictions, and provide guidance on where it should be used in WiFi deployments. In this first part, I'll discuss the immutable laws of radio frequency physics.
In radio communications, a receiver that hears multiple signals at the same time on the same (or overlapping) frequencies cannot distinguish between them. When this happens in WiFi, this is referred to as a collision. The multiple transmitters are causing co-channel interference (or adjacent-channel interference in the case of overlapping channels). The transmitted frame is corrupted, because the receiver cannot distinguish the intended frame from other signals on the same frequency.
In the WiFi protocol, a receiver sends an ACK to acknowledge proper receipt of the transmitted frame. If the transmitter does not receive the ACK, then it must assume a collision occurred and resend the same frame again (and again, and again) until the transmitter receives the acknowledgement. If a transmitter has to repeat the same frames over and over again, it diminishes the overall effective capacity of the channel.
When people complain about WiFi performance, it's a safe bet that quite a lot of collisions are occurring.
To provide contiguous coverage throughout a facility, some coverage overlap between neighboring access points is necessary. This can, however, createself-interference, where the APs within the same network cause co-channel interference and result in collisions for client devices in these overlap zones. To avoid self-interference, WiFi engineers can apply a process calledchannelization, which involves setting independent channels on neighboring access points and carefully controlling their individual transmission output powers so that their signals do not interfere with each other in these overlap areas.
 (Image: Pobytov/iStockphoto)
The ease or difficulty of channelization depends on the number of independent channels available. Since the size of the usable frequency band for WiFi per regulatory domain (i.e. country) is fixed, mathematics dictates that when the size of each independent channel is doubled, the number of available independent channels is halved, rounded down. Thus, as channel sizes get larger, channelization gets harder.
Larger channels also result in an increased noise floor, defined as the minimum threshold level below which a receiver cannot distinguish a signal from the background noise. There is naturally occurring background energy called thermal noise, as every object in the universe gives off very low levels of energy that are a function of temperature and bandwidth. For 20 MHz WiFi channels, this floor is approximately -101 dBm.
However, as channel sizes are doubled in size, the level of background energy also doubles (i.e. rises by 3 dB), increasing the noise floor to -98 dBm for 40 MHz channels, -95 dBm for 80 MHz channels, and -92 dBm for 160 MHz channels. Since the achievable data rate between two radios is a function of the signal-to-noise ratio (SNR) at the receiver, a higher noise floor results in lower effective maximum transmission speeds.
Most client devices, especially smartphones and tablets, have much weaker transmitters than the access points. Accordingly, a WiFi network really must be designed to maximize the signal strength of client devices, as received at the access point. The only practical way to increase SNR, therefore, is to minimize the distance and obstacles between the access point and the client devices. This, in conjunction with the growing need to support more and more client devices on the WiFi network, leads to deploying more access points to cover a given area.
Unfortunately, adding more access points can actually decrease the total capacity, since more access points and fewer channels makes channelization harder and self-interference more likely. Proper channelization of the access points becomes increasingly important. While counter-intuitive, larger deployments with more densely placed access points generally have higher total capacity with smaller channel sizes and proper channelization.
Below, I'll discuss how the laws of RF physics intertwine with IEEE standards and worldwide regulatory domains to make the use of bonded channels challenging.

Channel Bonding In WiFi: Rules And Regulations

(Appeared in Network Computing on 1/25/2016)

The second part of this blog series examines the IEEE standards and government regulations that impact channel bonding.
Channel bonding has become a very useful mechanism for accommodating growing WiFi data capacity requirements. But as I explained in my last blog, the laws of radio frequency physics can make channel bonding challenging to use in practice. In this post, I discuss the impact of IEEE standards and regulatory restrictions on channel bonding, and how this makes the use of channel bonding even more complicated.
In the IEEE 802.11a and 802.11g WiFi standards, channel widths were strictly defined as being 20 MHz in size.  The number of independent channels varied by country, but most regulatory domains allowed for at least three channels on the 2.4 GHz band (802.11g) and at least five channels on the 5 GHz band (802.11a). Channel bonding was first introduced with 802.11n to allow 40 MHz channels, and then ultimately extended further with 802.11ac to allow 80 MHz and 160 MHz channels.
As mentioned in my last blog, the size of the usable frequency band for WiFi per regulatory domain (i.e. country) is fixed. Accordingly, mathematics dictate that when the size of each independent channel is doubled, the number of available independent channels is halved, rounded down.  
While simple enough to understand, the laws of mathematics wreak havoc on the 2.4 GHz band when channel bonding is applied. The usable frequency band for most of the world on the 2.4 GHz band is only 72 MHz wide (2.401 GHz to 2.473 GHz). Accordingly, there are only three independent 20 MHz channels, and it is mathematically impossible to have more than one independent 40 MHz channel.

In contrast, the 5 GHz band has up to 500 MHz in the US, providing up to 25 independent 20 MHz channels, as opposed to only three independent channels on the 2.4 GHz band. I say “up to,” however, because this band is only semi-contiguous; many regulatory domains and the historical evolution of FCC regulations of this band make it much more complicated to use, especially when bonding the channels together.
Nearly half of the frequency range of the 5 GHz band requires dynamic frequency selection (DFS), a process by which the access point must detect the signature of existing government weather radar and other radio systems and vacate the channel for an hour. If a particular WiFi network is located in range of such an existing radar system on a DFS channel, the particular DFS channel is unusable at that location, and channelization must be redone to avoid that specific channel.
In March 2014, the Federal Communications Commission (FCC) extended DFS detection requirements to apply to client devices as well. Unfortunately, many WiFi consumer device and appliance manufacturers, driven by cost and time-to-market constraints, took the easy way out: simply do not operate on any channels in the DFS channel range. This significantly limits the practical number of channels available if those client devices are on the network, especially when channel bonding for larger channels. But don’t worry: All of your Apple and virtually all of your Android smartphones and tablets currently support operation on DFS channels. The problem is with low-end laptops and wireless USB adapters, which will revert back to 2.4 GHz operation if they cannot see a 5 GHz signal because the AP is using a DFS channel.  
The following table shows the full range of 5 GHz channels for different channel sizes.

  • UNII-1 (Channels 36 – 48, 5.170 – 5.250 GHz):  This 80 MHz wide band historically was regulated by the FCC for low-power indoor use by WiFi, but is currently allowed for both indoor and outdoor WiFi at the same power levels as all other 5 GHz channels. Most regulatory domains worldwide allow WiFi operation in this band. All 802.11n and 802.11ac client devices can use this band.
  • UNII-2 (Channels 52 – 64, 5.250 – 5.330 GHz):  This 80 MHz wide band historically was reserved by the FCC for government weather radar systems, so DFS sensing is required by access points and client devices. Most regulatory domains worldwide allow WiFi operation in this band. Only higher end 802.11ac client devices and older 802.11n client devices support use on this band.
  • UNII-2e (Channels 100 – 144, 5.49 – 5.730 GHz):  This 240 MHz wide band also was reserved by the FCC for government weather radar systems, so DFS sensing is required by access points and client devices. To complicate matters further, Channel 144 was only added for WiFi use in 2013, with the emergence of 802.11ac, in order to support an additional 80 MHz channel. Hence, older 802.11n client devices and some access points do not recognize and therefore cannot operate on Channel 144. Furthermore, several worldwide regulatory domains don’t allow the use of this band. Only higher end client devices and older 802.11n client devices support use on this band.
  • UNII-3 (Channels 149 – 161, 5.735 – 5.815 GHz):  This 80 MHz wide band was historically designated by the FCC for both indoor and outdoor WiFi use, though several worldwide regulatory domains do not allow WiFi in this band. All 802.11n and 802.11ac client devices support use on this band.
  • ISM (Channel 165, 5.825 – 5.845 GHz):  This 20 MHz wide band is also allowed by the FCC for WiFi, though. several worldwide regulatory domains do not allow WiFi in this band. All 802.11a, 802.11n, and 802.11ac client devices can use this band, though this band is not used when larger channel sizes are deployed.
The FCC is considering the addition of additional 5 GHz spectrum, including the 5 MHz between the UNII-2e and UNII3 bands as well as adding additional channels (i.e. 68 – 96 and above 165). Given that there are more government systems currently using those frequencies and the general speed of decision making, do not expect to see any further expansion of the channel space in the next three to five years. Even if additional spectrum is made available, DFS or similar restricted use of these expanded frequency bands will likely apply.
So where can you actually use these larger bonded channels?

Channel Bonding In WiFi: Use Cases

(Appeared in Network Computing on 2/15/2016)
Bonded channels make sense in certain types of WLAN deployments, but not others.
In this series on channel bonding in WiFi, I've discussed bonded channels from a technical perspective, both in terms of the radio frequency physics as well as the regulatory domains and standards. In this final post, I'll provide guidance as to where -- and how -- you should use bonded channels.
As with all engineering efforts, the choice of using bonded channels depends on the requirements (i.e. what level of service do you need) and constraints (i.e. what factors must you live with) of your WiFi network. Bonded channels are more appropriate when used in WiFi networks driven by WiFi coverage, such as those in private homes and s small or medium businesses, as opposed to WLANs driven by WiFi capacity like those  in large venue environments.
I discussed how two APs on the same or overlapping channels create co-channel interference, which prevents frames from being received, resulting in retransmissions and a lower total channel capacity. Co-channel interference can occur from either your own APs (i.e. self-interference) or from external neighboring WiFi systems. While the effects of co-channel interference are independent of the original sources of that interference, there are two distinct practical differences when designing and deploying WiFi networks:
  • In networks consisting of multiple APs, most interference is self-interference (i.e. from your own APs). This should be readily apparent, as most of the immediate neighboring APs will be internal to the same network.
  • While you cannot control neighboring WiFi systems, the access points within your own network are under your own control. Thus, self-interference from your own APs is largely avoidable with proper design and deployment.
One can get away with a single 40 MHz channel on 2.4 GHz and a single 80 MHz channel on 5 GHz in a small to medium size private home with neighbors that are not close enough to cause interference. Hence, consumer wireless routers are suitable for this environment. Unfortunately, this starts to break down in larger private homes, where multiple access points need to be deployed. In such environments, 20 MHz channels must be used at 2.4 GHz, though here the driving requirement is coverage and not capacity, so 80 MHz channels on 5 GHz can usually be used.

Most WiFi vendors advertise “auto channel” or “radio resource management” (RRM) features to perform channelization automatically. This is clearly convenient, especially as the number of access points grows into the tens, hundreds, or even thousands. However, such adaptive algorithms, no matter how elaborate, have inherent limitations, especially with a small number of independent channels. A channel change on a single AP can ripple throughout the entire network, causing service disruptions and creating new sources of co-channel interference, which will lead to further AP channel changes, and so forth. It's generally best not to rely on such automation, but instead establish a static channel plan to avoid self-interference.
WiFi networks consisting of more than two access points require a minimum of three independent channels to control self-interference. That said, some level of self-interference is generally unavoidable with only three channels, especially as the density of access points increases to add additional network capacity. The more independent channels there are to work with, the easier it is to minimize or even simply avoid self-interference.
As mentioned above, the 2.4 GHz band has only three independent channels at 20 MHz, and thus only one independent 40 MHz channel can be created. Accordingly, for the 2.4 GHz band, it is never appropriate to use 40 MHz channels in any multi-AP environment. On the 5 GHz band, 40 MHz is typical for a wide variety of deployments. Given the current channel allocations, using 160 MHz is impractical in any multi-AP deployment, and 80 MHz channels tend to be practical only in relatively small deployments (on the order of up to 20 – 25 APs) where DFS channels are usable.
The “consumer side” of WiFi can generally be broken down into small private homes, large private homes, and multi-dwelling units such as  apartment buildings and student dormitories. WiFi tends to be supplied by consumer-grade wireless routers designed to cover a 2,000 to 2,500 square-foot area, either purchased from a consumer electronics store or supplied by bandwidth providers such as telco or cable companies.
These wireless routers often default to using 40 MHz channels on 2.4 GHz and 80 MHz channels on 5 GHz, and often don’t even provide the “nerd knobs” in the configuration GUI to change them. On the 5 GHz band, generally only the UNII-1 (channels 36 – 48) and UNII-3 / ISM bands (149 – 165) are available, meaning that there are only two 80 MHz channels available on the 5 GHz band.
Where bonded channels become really problematic is in multi-dwelling units, such as apartment buildings, assisted living, and student housing. Consumer wireless routers are generally designed to cover a 2,500 square-foot private house. When they are placed in an apartment larger than 1,000 square feet, the signal bleeds over into several surrounding apartment units (both horizontally and vertically), each of which generally has its own consumer wireless router. It is a testament to the robustness (or perhaps stubbornness) of the 802.11 protocol that WiFi in such environments can even work at all.
In these environments, therefore, it is more appropriate to install a centralized WiFi network using enterprise-grade access points that are not necessarily deployed in every single unit. Deploying the APs in hallways is convenient, and sometimes unavoidable, from a cabling perspective. However, the hallway itself will acts as a focusing lens for the RF signals, which can result in self-interference with APs that may be quite a distance apart from each other.
Additionally, the strongest signal coverage winds up being in the hallways, when you really want the strongest coverage to be in the units. Most apartment units put bathroom mirrors, metal appliances, and other structure on the hallway wall, which all serve to attenuate RF signals and thus decrease the quality of the coverage inside the unit itself.

channel bonding uses 1.jpg

Predicted signal level (left) and self-interference (right) for a hotel with APs installed in hallways.
Consequently, it's better to place individual access points inside every few units in a staggered pattern both horizontally and vertically. While capacity is an ever-increasing concern, the WiFi requirements for multi-dwelling units are still oriented towards providing coverage vs. capacity. Hence, 40 MHz channels on the 5 GHz band are typical, and in some environments even 80 MHz channels on the 5 GHz band are usable.

channel bonding uses 2.jpg

Predicted signal level (left) and self-interference (right) for a hotel with APs installed in rooms.
For high user density environments, such as conference centers, stadiums, arenas, shopping malls, and lecture halls, the requirements are driven by capacity, as there are a large number of devices that are connecting in a relatively small area. In these environments, multiple APs are installed, usually with highly directional antennas, to segment the area into as many parallel sectors as possible to handle hundreds or even thousands of simultaneous users. There can be a large number of overlapping access points, so typically 40 MHz or even simply 20 MHz channels on the 5 GHz band are used.
Overall, channel bonding is a very useful feature of WiFi to increase capacity, but applicable to particular environments. Without carefully controlling channelization and channel sizes, simply adding more access points isn’t always better, and in fact can make the network worse.

No comments:

Post a Comment