Continuing the discussion of the budget muni wireless network requires analyzing the front line component, the Access Point (AP). There are many variations of APs. They all have features and capabilities that provide enhancements in certain environments. Some of the APs break the practical limit of 20-30 users by using multiple radios in a single enclosure, proprietary polling systems (not compatible with other vendors), and advanced beam-forming techniques. However, the focus is from a budget standpoint and that means this design will start with a single 2.4 GHz radio with an omni-directional antenna. In later articles, I will cover upgrading the design to support more users and a larger coverage area. The beauty of this design is that it’s cheap to get in the game and scalable.
The ideal inexpensive AP was covered in the first article of this series. This article discusses the protocol it should support. Obviously if it supports 802.11n, that’s a big improvement. 802.11n can provide up to twice the range of 802.11 b/g with the same single AP, single antenna design. This is due to a combination of a more efficient transmitting modulation scheme and better receiver design than 802.11b/g/n. In 802.11n, this would be considered a 1×1 MIMO. If the world was perfect and everybody had 802.11n devices, theoretically it would take only ½ to ¼ as many APs per square mile for the basic system. Unfortunately, the expansion of WiFi phones and their lack of support of 802.11n means keeping legacy 802.11b/g compatibility, range, and load in mind. If the network doesn’t need to support 802.11b/g, the savings are significant. The reality is though, if the network was built today for the general public, the system should still support 802.11b/g equipment because many people are still running around with laptops, smartphones and other devices that have only 802.11b/g wireless chips.
Taking the AP design a little further, figure out what the client connectivity area is. If the goal is to connect to Joe or Jill Teenager, who lives in suburbia with maple trees covering the sky and every home is built like the Windsor Castle (brick), while he or she is lounging on the couch with an iPod Touch which doesn’t support 802.11n, then even with 16dBi omni antennas, the AP will have to have to be within a couple hundred feet. If they live in Stuccoville, Arizona, (yes, stucco attenuates signal, but go with me on the flora thing) the range will be much greater since trees are 60 feet tall with leaves starting at 55 feet (palm trees). The only other vegetation is 2 feet tall and is considered a lethal weapon (cacti for the Northerners). This is a slight exaggeration, but the basic physics apply. If the clients are outdoor only such as surveillance cameras, mobile hot spots for police, utilities, etc., then the number of APs can be reduced significantly. In fact in Arizona, there are places where 2 APs per square mile are all that’s needed for mobile coverage or every traffic light corner.
Lessons from a small Wi-Fi deployment
The really interesting thing concerning small deployments such as public parks and mobile home parks, is that the back end Internet bandwidth options are rarely as fast as the radio capacity for cost reasons, access, or the restrictions in end-user license agreements. For example, one park that Triad manages is really only under heavy use for two months during the year (Spring Training) in addition to 2 to 4 high-use specialty events. Bandwidth options are a T-1 for a whopping 1.5Mbps ($450 monthly), DSL for 12Mbps ($100 monthly), or cable for 4Mbps ($200 monthly). Assuming 12Mbps is enough (it can be expanded with multiple DSL circuits as needed), users are limited to 2Mbps, the radios support up to 20Mbps in 802.11g, and there are only 2 hops maximum meaning no need for multi-radio APs. Even at 2 hops, bandwidth is still 10Mbps on the perimeter. There are 11 radios total to cover about ½ square mile with 2 high-density cluster areas. The 2 high-density areas are each comprised of 4 separate radios at the perimeter locations. The remaining areas are covered by 3 other radios. Using high-gain omni-directional antennas, laptops can connect at 800-1200 feet. All radios are in AP+WDS mode for relay and coverage. The total cost for all the radio equipment and antennas for this deployment was $2300, not counting the tower mounting brackets. The system has been up for over 2 years and is scheduled for an 802.11n upgrade in the next month or so. The omni-directional antennas will stay, meaning that the 2 hop system will now deliver about 20-30Mbps on the second hop.
The project utilizes four vertical assets up to 150 feet that also include a 5.8GHz PTMP system. The 5.8GHz PTMP system has a range up to several miles in a 360 degree pattern. Each vertical asset consists of one radio with a standard vertical polarity sector antenna. When the PTMP system is upgraded, it will consist of 5.8GHz 802.11n, 2×2 MIMO equipment. The cost of the upgrade will be approximately $250 for the radio and antenna and will support up to 100Mbps per radio. Throw in another $100 per pole for backhaul and the system has now expanded from a simple hot-spot project to a 100 square mile 400Mbps infrastructure for an additional $1400.
Secrets of 802.11n
It’s time to step back and discuss the unadvertised secrets of 802.11n since it affects the expectations of the design. Here’s a big shocker – the processor and firmware of the AP affect the radio performance. For example, the AP manufacturer claims 300 Mbps. That’s modulation, not real-world throughput. That number is also rounded up from the real number which is between 270-288 Mbps, depending on what’s called the guard scheme (not within the scope of this article and makes my head hurt). Keep in mind that many of the 802.11n radios also have 100 Mbps Ethernet jacks because manufacturers know that communication is half-duplex. The Ethernet port is full-duplex so they consider 100 Mbps up and 100 Mbps down, 200 Mbps total, not 200-300 Mbps in one direction. Strike one.
The second issue is that 300 Mbps is only achievable by running 40 MHz wide channels. That works great in the living room for 50 feet. It doesn’t work so well when the AP is perched on a light pole with 200 houses in range. It would be difficult at best to go 500 feet in 2.4GHz and get maximum theoretical modulation rates with a 40 MHz wide channel. Multi-radio APs typically use the 5 GHz bands for backhaul for that reason. That means the real-world useable 2.4 GHz bandwidth is 20 MHz which translates to 150 Mbps. Strike two.
The 5.8GHz band is the most commonly used with 40 MHz channels. 5.1 GHz to 5.3 GHz is usable with some manufacturers, but the EIRP drops significantly. However, for 500 feet and no vegetation, that is reasonable. There should be very little interference in those bands. Of course, many of the radios that are already in those bands are probably set up by a local WISP. Although most WISPs are knowledgeable and legal, there are a few WISPs that either have no clue about the rules or are intentionally broadcasting illegally due to congestion in the 5.8 GHz band. There are manufacturers who have certified equipment in those bands, but because of the limited EIRP in those bands, the equipment isn’t as popular. 5.8 GHz also isn’t supported by most laptops and smartphones (note: Apple’s Airport Extreme wireless base station, as well as the latest iMacs and Mac Book Pros, support 5.8 GHz).
Assume that there isn’t interference in whatever band with a 40 MHz wide channel. Then the next question is how far can a 300 Mbps modulation level be maintained? Well, here is the second problem. To quote a good friend of mine, “speed, distance, reliability, cost – pick 3”. The first 2 are the most critical. For just the radio, it should be “speed, distance – pick one”. Basically, to get the 300 Mbps modulation rate, receiver sensitivity drops to around -72- to -74dBm and the power output of the radio drops from the fabled 26-30dBm to around 24-26dBm or less for outdoor equipment and 15-18dBm for the retail $30-$200 equipment. To get the magical 300 Mbps speed in a 360 degree coverage pattern from a single radio with an omni-directional antenna, there should be total LOS, no interference, and at maximum distances of around 500’ or less with the 40 MHz channel. Most multi-radio AP manufacturers recommend directional antennas but one could argue that defeats the concept of ubiquitous mesh architecture. Strike three.
I have not done testing with most of the new 802.11n APs so some of you can jump in here with some real-world values. I’m also not going to get into sector antennas here for pole installations because the size of the antennas which makes it difficult to get through city zoning. I’m going to also get some grief here from the beam-forming guys; but realistically, none of the beam-forming systems in 2.4 GHz can match a 28-inch tall sector antenna in performance. It’s a simple matter of physical capture area. However, newer sector antennas also support multi-polarity which gives them even more advantage.
Although baseball doesn’t have a strike four, here it comes. 300 Mbps modulation rates are a real-world throughput of 150 Mbps through one radio under absolutely ideal conditions. This typically doesn’t exist in most suburban realities. In addition, the processor is also a bottleneck for the AP. On some 802.11n devices, a straight FTP transfer from one computer to another computer with a 20 MHz wide channel (much more realistic) will result in a transfer of about 35 Mbps. Put two computers on each side, and then do the same transfer computer to computer, and the throughput jumps to about 60 Mbps with each stream about 30 Mbps. Do a 3×3 transfer, bandwidth goes up to 72 Mbps, and each unit drops to around 24 Mbps. Further testing up to 10×10 transfers demonstrates 85 Mbps maximum. All of these numbers are perfect signals at short range in the lab. So what happened?
Basically, some processors/chipsets get more efficient as multiple transfers of data occur but are fairly limited for a single transfer. For example, using a 20 MHz channel and MCS7 modulation rates (65-72 Mbps depending on guard scheme), will result in a 60 Mbps with 65% CPU overhead with UDP video traffic from a single camera. However, change that to TCP/IP and the rate drops to 35Mbps and CPU processor overhead jumps to 100%. This test was done with a 400 MHz Atheros processor and chipset in the radio. Other manufacturers also use the 300 Mbps but really add in the combined throughput of 2 radios, not a single data transfer rate. Keep in mind that this isn’t for every single product out there, but it shows that real-world testing is definitely required before a design is signed off on for the particular application.
802.11 a/b/g APs exhibit this same behavior when compared to the marketing material so it’s not a new phenomenon. In some cases, manufacturers used UDP traffic numbers instead of TCP/IP real-world traffic. In other cases, manufacturers were using 40 MHz channels for their specifications which again, aren’t realistic for most installations. In many cases, when 40 MHz channels were used, the CPUs capped out so that the result was not a doubling of the transfer rate but maybe 80% more over a 20 MHz channel. For example, 20 Mbps at 20 MHz translated to 35 Mbps at 40 MHz.
Another problem with processor overhead is how many packets per second (PPS) that can be jammed through at one time. A user opening up a web-site usually has a fairly low PPS requirement, thus low CPU overhead. Open up a file-sharing application and the number of sockets and PPS can go through the roof. Low-cost APs trying to handle this kind of traffic typically slow down drastically. More expensive and faster processors along with better firmware scale better under load.
There are many reasons that one AP costs $100 and another costs $6100. More expensive APs will use 2, 3, 4 or sometimes more radios within a single AP enclosure. The rest of the costs come in the form of firmware, management tools, and capabilities. Many of the outdoor units have firmware designed to optimize video transmission quality, fast-handoff between AP radios while vehicles are moving, mesh implementation, beam-forming, multiple SSIDs, multiple frequency, channel bonding, and many other advanced WiFi technologies. 802.11n radios are also start with a better specification foundation.
Switching back to the budget system, WDS communication is built in to almost all the chipsets and supported in firmware. This allows APs to connect to each other through Layer 2. It’s simple, sometimes not compatible between manufacturers, and requires manual setup with MAC addresses of each device. Mesh firmware, with a basic setup, simply finds all other units in an area, then sets up either a layer 2 and/or a layer 3 network between the radios dynamically, as part of its underlying mapped infrastructure. Mesh firmware is always proprietary between APs.
The goal of Tales from the Towers is to create a budget, scalable, municipal WiFi system. It will probably never have many of the capabilities that more expensive systems can do. The budget WiFi system will not initially and probably may never support fast handoff for moving vehicles, optimized video traffic, or possibly multiple SSIDs. Therefore it’s important to understand what its capabilities are before deciding that price is the most important factor. In some cases, its inexpensive nature, flexibility, and scalability allow the network to get around many of the advanced features that some of the more expensive APs have. In others cases, it simply isn’t going to happen and the proper design should be Motorola, SkyPilot, Firetide, Tropos, MeshDynamics, Ruckus, BelAir, Meraki, Meru, or whatever product fits the needs and budgets of the municipality. Every one of these companies has features and benefits that make them ideal for specific applications. I have designed and installed systems using many of these products.
However, the focus here is a starter system the meets the cost and scalability of really tight budgets. This makes it deployable in areas that can’t justify the ROI of more expensive equipment and can bring Internet to areas that may not be able to justify a large investment. The next article will cover the actual AP deployment design.