Not As Fast As You Think…..

As you might imagine, we spend a lot of time here at Jarvinian testing all things wireless. The antenna tests mentioned in an earlier post are just one example. We also regularly test the cellular networks in our area and when we travel. Each of us has a few apps installed on our phones designed to test the speed of cellular networks and while none seemed particularly accurate in terms of actual speeds, they all provide some insight into the relative speed of a network in different locations. However, we recently found an app that returned absolute network speeds more in line with what know to be reality.  The story of this app reveals a lot about the public’s (mis)perceptions when it comes to the speeds of wireless networks.

The app in question is Cell Phone Coverage Map by Rootmetrics. This app has a number of unique features, including creating a coverage map based upon tests performed by other users of the app. And while the crowd-sourced coverage map is interesting, the thing we liked best about the app was the accuracy of the speed test.

The Rootmetrics Coverage Map

Based on our own independent tests, the results of Rootmetrics speed test were by far the most accurate of the speed test apps we have used. Unfortunately for Rootmetrics, actual network speeds are generally much slower than users expect. Conditioned by the marketing departments of cellular providers, many cellphone users assume that the speeds they are seeing on their networks are near the capability of the standard or at least are a significant fraction of the capability. According the to published standards, 3G for either standard (GSM or CDMA) has the “capability” of speeds well above 10 Mbps. Of course, the capability depends a great deal on the modulation scheme being used, which in turn depends on the noise environment. What that means is that while 10 Mbps is certainly possible in a quiet lab environment, it’s much harder to achieve these speeds in the real world.

The thing most people don’t understand is just how much the chaotic environment can degrade network speed. For people who care to test their cellular network, downlink speeds of 1 to 3 Mbps appear reasonable if somewhat disappointing. After all, that’s only 10-30% of the capability of a 3G network. Unfortunately, the reality is that a combination of factors can combine to degrade network performance to speeds closer to 1-3% of the capability. Interference and noise introduce errors, which in turn force the network to use less efficient modulation schemes and can drastically reduce the network speed. An accurate test of network speed will reflect this reality.

The Cell Phone Coverage Map app did this quite well. Which turns out to be great if you wanted accurate results, not so great if you wanted people to download your app. Looking at the reviews of the app, people simply couldn’t accept that the slower speeds were real. One reviewer gave the app one star and wrote, “for whatever reason, when testing 3G speeds, results posted are between consistently 1/10 – 1/5 the actual speeds experienced.” Another one star review said, “This app reports low numbers all the time regardless of location.” A third bad review said this: “This app consistently reports speeds that are much, much lower than any other bandwidth testing app out there.” Finally, another one star reviewer wrote, “I did a speed test using this app and got 400kbps. I did a speed test with 3 other apps and got speeds of 2000 – 4000 kbps.” None of these reviews provide reasons as to why they think the slower speeds are incorrect – they just assume they are wrong. When faced with the possibility that the slower speeds are correct, many users refuse to believe it. Despite other testing apps showing higher speeds, based upon what we know about the current state of cellular networks, it is highly likely that the slower speeds reflected in the Rootmetrics app are much closer to reality.

Unfortunately, from what we can tell, the folks at Rootmetrics have responded to this by modifying their app so that the tested speeds are in line with other competing apps. While this may not be accurate, it keeps them from being an outlier and increases the likelihood that users download the app.  This in turn means more users providing data for their coverage maps, which are clearly the focus of the company.

So we’ll keep looking and running speed test apps through their paces. Maybe someday someone will design an app that provides continual network testing in real time, allowing users to see clearly how quickly and dramatically network speeds can change in response to the chaotic wireless environment. We can only hope.

Wireless Handsets: Moving from Fluff to Real Stuff

The name of the game in the market for wireless handsets is about to shift from features and functionality to performance. Handset providers that recognize this shift in buyers’ decision-making and act proactively to capitalize on it stand to leave their competitors in the dust.

Competition in the wireless handset market to date has centered on who provides the most features and functionality – the coolest style, the most apps, the best camera, the largest screen, or the best operating system. In order to sell more handsets, providers have focused on phone features and functionality to the detriment of performance – having to sub-optimize components to make handsets slimmer, lighter and with more bells and whistles. Apple decision to put the iPhone 4’s antenna on the outside of the case resulting in poor performance is a good example of this (see Chris Carter’s post, Antenna Madness, for a more thorough discussion).

There is an increasing lack of differentiation in handsets today. New handsets today are beginning to look the same and offer similar features and functionality. There don’t appear to be any new killer apps or features on the horizon. How can a handset provider distinguish themselves to stand out from the pack?

Although consumers care about handset performance, due to the marketing wars between network providers, they mistakenly believe that the networks are the primary determinant in the performance of their handsets. Surprisingly there is no focus on the performance characteristics of the phone itself. Handsets are a critical part of the overall performance of the mobile network. Small changes to the performance of a handset can have a significant impact on the overall performance of the network. More efficient handsets send data and information with less need for error correction, which means less bandwidth and battery power are used. This in turn improves the performance of the handset and the network.

As the demand for spectrum explodes, further burdening already overtaxed networks, and with less differentiation among handsets, we believe consumers will shift their buying preferences from handset features and functionality to performance. Handset performance will become the primary determinant in market success. The handset providers that recognize this shift and change their marketing messages to focus on performance will out distance their competitors that don’t.There is a need to establish objective performance benchmarks or metrics so that consumers can compare handset performance on an apples-to-apples basis. Currently there are no objective benchmarks that measure handset performance. It would be useful if an industry group, such as the FCC or CTIA, were to establish common performance benchmarks that all phones could be tested against. We have some thoughts on the metrics that ought to be used to measure handset performance, but will save that for a later post.

Signal, Noise, and Speed

People often ask us at Jarvinian about the rather dramatic differences between the published data speed of wireless networks and the actual (much lower) data speeds experienced by users.  In a nutshell, our answer is SNR or Signal to Noise Ratio.

All wireless networks must cope with the scare resource that is radio frequency (RF) spectrum.  An average network must service millions of bandwidth consuming users with a range of frequencies that is an infinitesimal fraction of that which can fit into a single optical fiber.  Thus, to meet exponentially growing demands for data throughput speed, wireless networks must achieve fantastically high levels of spectral efficiency.

The most potent tool for achieving high spectral efficiency is advanced modulation, and 4G standards like LTE employ signal modulation protocols that can theoretically squeeze a massive amount of data into a comparatively narrow range of radio frequencies.  However, there is a catch.  As modulation becomes more spectrally efficient, it also becomes more susceptible to degradation from RF noise and interference.

The figure above shows a “phase constellation” for 64QAM, a highly efficient modulation method used in 4G applications.  Each “phase” is subtle change to the signal carrier that represents a string of binary 1′s and 0′s.  As we add more and more discrete phase states (in this case there are 64), the theoretical spectral efficiency rises.

On a relatively noise free wireless channel, this system works very well and can produce extremely high data speeds. However, on a wireless channel with high levels of noise and interference, the many subtle phase relationships become difficult to distinguish and the effective data throughput speed plunges rapidly.

Real world wireless channels are often full of noise and interference, and this is the reason why next-generation networks often fail to achieve their advertised data speeds.  As more and more network activity crowds the spectrum, we can anticipate that the level of noise and interference will continue to rise, effectively lowering the data throughput capacity of mobile infrastructure.

To remedy this situation, we believe that the mobile industry needs to get serious about innovation in some very long under serviced areas of wireless technology.  From our perspective, the future will not be about more spectrum (there won’t be any). Rather, it will be about “conditioning” existing spectrum to be noise free enough to support the data throughput potential of advanced modulation.

Antenna Madness

On its most recent earnings call, Verizon hinted that the next iteration of the iPhone would work on both CDMA and GSM networks. Right now, the Verizon iPhone only works on CDMA networks and the AT&T iPhone only works on GSM networks. Combining the two into one phone certainly makes sense for Apple – manufacturing one model of iPhone instead of two creates manufacturing efficiencies and simplifies logistics. But it will probably make life a little harder for both Verizon and AT&T. Since GSM and CDMA networks operate on some different frequencies, the iPhone 5 will have to be able to handle all of these frequencies.

As mobile networks have evolved, so have the phones that work on those networks. Given the complexity of spectrum allocation plans in countries around the world and the high cost of spectrum, wireless providers have been forced to acquire spectrum where it was available rather than where it was best suited to their needs. When a provider network utilized only one set of frequencies, single band phones were acceptable. Each new set of frequencies acquired by the providers required that the phones be able to operate on the new frequencies in addition to the old frequencies. Consequently, over the past fifteen years, handsets have evolved from single band through dual band and tri band phones to the quad and penta band phones of today. Some of the biggest providers in the world today are already planning for a future with 10-band and 11-band phones.

The general perception is that getting a phone to work on another frequency simply requires additional software and perhaps a new chip or two in the phone. And while new software and chips are often required, looking at it this way ignores the critical importance of the antenna. As Apple learned with the iPhone 4 “Antenna-gate”, getting the antenna right is critical to the proper functioning of a phone. Unfortunately for phone designers, when it comes to antennas, the difference between an electromagnetic wave at 700MHz and one at 2.5GHz is substantial. It turns out that one size does not fit all.

To get a better understanding of this, it is helpful to go back to the days when cell phones had external antennas. When the phone only had to receive and transmit on one frequency, the phone designers could add an antenna that was resonant with that frequency. Take for example the old Motorola phone pictured to the right.
The whip antenna on the phone extended to a length of about 9 cm, which is about ¼ of the wavelength of an electromagnetic wave at 835MHz.
In general, antennas have better resonance with radio waves of a specific frequency when they are made in lengths that are a whole fraction of the wavelength of that frequency. So in the case of this phone, having an antenna size that is about ¼ of the wavelength of the one frequency the phone used to receive and transmit was very effective.

In fact, we did some experiments with the whip antenna from one of the old Motorola phone to see exactly how the antenna performed. Below is a graph of the frequency response of the whip antenna.

As you can see, the performance of the antenna is the best in the range of frequencies from about 720MHz to about 900MHz, with a fairly sharp drop off in response on either side of that range. This antenna did basically what it was supposed to do.

Now let’s look at the iPhone antenna. The iPhone 3G was a “Quad band GSM/GPRS/EDGE” phone so it had to receive and transmit at 850, 900, 1800, and 1900 MHz. The phone was also a “Tri band UMTS/HSDPA” phone which required it to operate at 2100MHz in addition to the other frequencies. That’s a total of 5 frequencies ranging from 850MHz to 2100MHz – quite the challenge for antenna designers. What did they come up with? The antenna pictured to the right which was located inside the phone at its base.

A far cry from the antennas that most of us are familiar with. How well did it work? It was okay. We did the same test with the iPhone antenna that we did with the whip antenna to see how the antenna responded over the same range of frequencies. The results are below.

The overall response was higher, but the antenna did a poor job differentiating frequencies. In the critical range between 700MHz and 900MHz, the antenna’s response swung wildly. Unlike the whip antenna where the best response occurred near the desired frequency and fell off rapidly on both sides, the iPhone antenna had a very high response at frequencies lower than desired and the drop off was relatively slow in the higher frequencies.

An antenna like this does a poor job of picking out the desired signal from all of the noise that exists in the environment. With higher sensitivities to undesired frequencies, the likelihood of picking up interference is higher. This decreases the signal to noise ratio, increases the amount of error correction the phone has to do, and ultimately reduces the throughput speed of the phone. Translation – your download speeds get slower.

While we have yet to test the antenna on the iPhone 4, the problems Apple experienced at the launch suggest that the new external antennas are at best equal to the antennas of earlier iPhones and in some cases (i.e. when your hand bridges the two antennas) they are worse. Fortunately, there are indications that Apple is taking antenna design more seriously now. But for Apple, this was a short-term problem. As “Antenna-gate” fades into history, users have gone back to blaming the network for poor phone performance rather than the phone. This is a positive for Apple, but a negative for wireless providers around the world.

Unfortunately, as the new combined CDMA/GSM iPhone suggests, the problem is getting worse. As phones are required to work on more and more frequencies, antenna design gets increasingly complicated. The advent of LTE has the potential to complicate things even further, especially with the introduction of MIMO technologies.

While there is no single solution to the challenges antenna designers face, there are some technologies that can help. In future posts, we will discuss some of these technologies and explain how they can improve the situation. The good news is that small improvements in antenna technology can improve phone reception and throughput speeds, improve user perception of network performance and reduce infrastructure costs for wireless carriers at the same time.

Noisy Spectrum: A Dirty Secret

The FCC’s call to arms with their cry of “the looming spectrum crisis” in October 2009 certainly has focused the attention of the public and the press on the lack of radio spectrum available to support the expected demand for wireless bandwidth over the next decade. Despite an occasional misguided suggestion that the spectrum shortage is exaggerated, there appears to be wide recognition of the problem and the need to address it. Leaving aside the issue as to whether reallocating 500 MHz of spectrum will be sufficient to fix the problem as the National Broadband Plan suggests (we don’t think it will, but will save that for another post), there has been an astonishing lack of attention paid to the fact that different frequencies are better suited to address this shortage than others. The press, the public, the government and many in the industry seem to view all frequencies as alike in their ability to solve the pending spectrum crisis and that couldn’t be further from the truth.

Not all frequencies are created equal. It is well understood that different frequencies have differing propagation characteristics — lower frequency signals are challenging to control because they travel so far and that higher frequency signals don’t pass through walls or building easily. What doesn’t appear to be understood is that noise and interference varies significantly across the spectrum and that they have a deleterious effect on the spectral efficiency of the frequency (the frequency’s ability to carry information). The impact of noise and interference on spectral efficiency can be best understood if one imagines the difference between a conversation between two people sitting across the field from one another in an empty football stadium and that same conversation in a crowded noisy stadium. The only way to be heard in the crowded stadium is to speak very loudly (i.e. crank up the power), or to repeat what you are saying until the other person gets the gist of your message (i.e. error correction though re-sending data) – it isn’t nearly as efficient as talking in an empty stadium.

Noise and interference come from both natural phenomena (such as solar and galactic radio emissions and atmospheric absorption) as well as man-made influences. Spurious emissions from Doppler radars, antique paging systems that are still broadcasting, public safety systems, wireless networks and digital television broadcast signals can wreak havoc with not only adjacent frequencies, but with frequencies further up and down the spectrum (due to harmonic effects). As spectrum becomes more crowded in terms of the amount of activity in a particular band and the proximity of active bands to one another, undesired emissions from adjacent channels will increasingly spill over into a desired channel causing interference.

The following is a graphic illustration of the noise levels at certain frequencies over a 30-minute period in New York City.

The pink spikes represent active frequencies that include cellular uplink and downlink bands (698-763 MHz), frequencies for digital television (692-698 MHz), and public safety frequencies (763-775 MHz). The dark blue frequencies in the center of the graphic are in a relatively quiet area and include frequencies for radio astronomy, GPS and air traffic control – “mission critical” services which the government has tried to shield from noise and interference. In many cases, the green/yellow spikes that abut the pink spikes and the occasional small green spikes in the center represent noise and interference.

Although it is fairly obvious that noisy spectrum (aka “dirty” spectrum) wouldn’t be as efficient in transmitting information as clean spectrum, there appears to be a lack of analysis available that attempts to quantify that difference. We recently conducted some tests in the Boston area to determine the magnitude of the difference in the spectral efficiency of frequencies in the 700 MHz and, which reside in a busy area of the spectrum, as compared to frequencies in the 1600 MHz band, which are in a quiet neighborhood. Data from both frequencies was collected at several locations, inside and outside, throughout downtown Boston. We expected to find that the spectrum at 1600 MHz was more efficient than 700 MHz, but were blown away by the results – the spectrum at 1600 MHz was capable of carrying 4.5 times more information than 700 MHz. That means that a 2 MHz band of spectrum at 1600 MHz can potentially carry the same amount of information as a 9 MHz band of spectrum at 700 MHz.

With such a significant difference in spectral efficiency between two frequencies, it is critical to consider the effect of noise and interference in making decisions about spectrum re-allocation, acquisition and valuation. We intend to conduct additional analysis to quantify the effect of noise and interference across the spectrum and will report back as we learn more.