Since 1998, our staff has been involved in development of VoIP (Voice over Internet Protocol) hardware, mainly in the areas of performance verification, quality of service (QoS) and MOS testing.
VoIP promised to provide telephony services over the Internet at low or minimal cost, with speech quality similar to that of conventional telecommunications.
Regrettably, that promise has not yet materialized. Currently there are many VoIP users, including both individuals and mammoth organizations, who are disappointed and disillusioned about the quality.
Users in general expected quality similar to that provided by the mainstream telcos, the crystal-clear ITU-T G.711 telephony (48, 56 or usually 64 Kbps PCM), and its 32 Kbps ADPCM derivatives. Instead, they've found the quality, more often than not, to be similar to cell phones at their very worst — unusable or so close to unusable that they have to turn back to conventional services.
This white paper looks at what can and should be done to bring VoIP to fulfillment of its original promise, as a viable, reliable, low-cost alternative to conventional telephony.
TO BEGIN WITH THE INTERNET:
The Internet is an incredible network originally developed to support data communications among a select group of universities. But that Internet evolved, and e-mail became a major communications medium.
As traffic volumes increased, issues of reliability and availability began to arise, creating a growing need for routing diversity. The topology of the network rapidly evolved from a fairly linear arrangement to today's world-wide spider web of possible routes.
Where each of these strands of spider web intersects, there is a computer "traffic cop" or router, continuously monitoring all the strands of the spider web connecting to it. When a "packet" of data arrives, the traffic cop looks at the destination address, looks for an available network strand heading roughly in the right direction, and ships off the data packet in the general direction of the destination.
This "traffic cop" arrangement works just fine for surfing web pages or ftp-ing. If one or one thousand packets get lost or corrupted, the software simply asks the source server to re-send the last block of data or whatever, and things eventually work out satisfactorily and transparently to the Internet users.
The advent of VoIP places a new demand on the Internet's capabilities, for the very simple reason that person-to-person speech communication is more or less a real-time activity. This significant difference between computer data and human communications means that the Internet has little opportunity to buffer, request a re-send of data or other similar approaches used for data communications. This means that with VoIP, when the network has problems, VoIP will likely also exhibit problems.
VoIP promised telephony service over the Internet at low or minimal cost. The low cost is entirely dependent, however, on the use of relatively high-order speech compression. The trade-off, unfortunately, is that a high degree of compression radically increases the extent to which the data is vulnerable to any Internet condition that causes data loss or corruption.
Before any VoIP processing is applied, the original speech data begins with a 64 Kbps G.711 PCM signal. With VoIP, this signal passes through some complex mathematical processes that result in about a 90% reduction of the original data rate. With this amount of compression, each individual bit becomes more and more mathematically significant.
Although any "lossy" compression tends to reduce signal quality, at 90% compression, the VoIP signal can still qualify as acceptable to good.
The critical quality issue arises when even one bit of that compressed data stream is lost or corrupted, because that one bit actually represents about 10 bits worth of data.
Other kinds of compressed data transmission, for example, the compressed video signals used for direct-broadcast satellite (DBS) television, rely on forward error correction (FEC) to overcome adverse conditions and maintain good quality under most circumstances (read more about compressed video and FEC here).
In contrast, VoIP doesn't include the built-in resiliency that comes with FEC. As a result, when things go wrong, users may hear bursts of static or other noise. Also, when the performance of the VoIP system starts to deteriorate, speech intelligibility generally drops to a point where functional communication becomes an impossibility.
The question, then, is whether VoIP can produce the quality required for acceptable speech communications in a business environment. And in spite of the problems experienced in many applications to date, the answer is still a yes — it just takes some care in addressing the various elements that can contribute to poor network performance.
For example, one manufacturer with whom we've had dealings produced a gateway product that converted G.711 PCM telephony to the VoIP domain at 6.3 Kbps per channel. The intention was to give telcos or enterprises the ability to squeeze nearly 240 calls into a single dedicated T1 (1.544 Mbps) channel with a normal capacity to carry only 24 uncompressed 64-Kbps voice channels. A comparable increase in call capacity would also be achieved with a dedicated E1 (2.048 Mbps) channel as used in Europe and other parts of the world, which normally carry 35 64-Kbps voice channels. The speech quality was consistently good and the cost savings were significant.
Key to the success of this particular product was the use of dedicated end-to-end transmission facilities. The VoIP compression technology was effective, but the application relied on avoiding the unpredictable public-access Internet for inter-city transmission.
What about VoIP users in the normal Internet environment? Are there ways they too can achieve acceptable quality and good cost savings?
A SECOND LOOK AT THE INTERNET:
The publicly accessed Internet was not designed or intended to support business-grade, reliable telecommunications. Nevertheless, it is capable of delivering acceptable quality, though achieving that quality is often a matter of luck as much as planning. VoIP manufacturers and users need to understand the foibles of the Internet so that they can make deliberate choices to bias their operating set-up in favor of good quality.
The Internet can throw a variety of impairments at a VoIP signal that can often be serious enough to render meaningful communications impossible. These impairments include:
- Packet loss: One or more data packets gets lost on the network.
- Packet error: Some of the bits in the packet get corrupted.
- Phase jitter: Some bits in a packet, or the entire packet is difficult to decode correctly because of network synchronization problems.
- Packet reordering: Because of differential transit time delay between packets, packets arrive out of sequence with unpredictable results.
With VoIP and the public Internet, telephony or speech is divided into data packets, about ten per second. It doesn't take much by way of impairments to make speech difficult to understand or even to make VoIP totally unusable.
A different class of problem can impair the functionality of the transmission medium, as opposed to impairing the VoIP signal:
- Congestion-related impairments when an insufficient amount of data arrives to permit communications.
The public Internet is a wild place. Abuses such as spam, viruses, worms and denial-of-service attacks can overload networks and routers, causing normal traffic to grind to a halt. For a latency-sensitive application like VoIP, the effect of network congestion can be just as devastating to functional communication as impairments to the signal itself. Users and manufacturers need to seek ways to minimize the impact of congestion on VoIP to the greatest extent possible.
FINDING THE FORMULA FOR SUCCESSFUL VoIP:
Examining some of the major VoIP components making up the system will help users achieve the best value for an investment.
Remember that VoIP generally works well if everything is perfect — as if life were that easy! Also, different products may have been designed with different specific operating conditions as key priorities. Ergo, some VoIP products will likely perform better than others in your particular set-up.
CPE (Customer Premises Equipment):
Some services use a desktop appliance to connect with the Internet, such as an xDSL,cable or wireless modem device with a VoIP port included. If the service provider insists on the use of a specific product that happens also to be a low-performance product, your service is unlikely to meet expectations. It's important to understand what quality you can reasonably expect from such a device, and what alternative choices exist.
Some VoIP software relies on the user's PC sound card to provide coding and filtering. Some sound cards and PCs are not up to the task, and will provide disappointing results, regardless of the quality of network. Potential purchasers should verify just what is involved, and whether it is capable of providing the desired quality of service.
If the customer premises includes a shared LAN or local area network, LAN usage by others can cause VoIP problems or failure. Correcting this situation will likely require a different LAN design solution.
One key aspect of customer premises equipment is often overlooked, and that is echo. With echo, you get to hear your speech returned to you perhaps several tenths of a second after you speak, making effective, coherent speech really difficult. Echo control is built right into a conventional telephone network, but with VoIP, you're essentially creating your own network, and it's one that doesn't automatically include control of echo. One way to resolve the echo problem is for users at each location to use either a headset or a hands-free unit that includes an AEC (acoustic echo canceller).
The Last Mile:
In telecommunications, the last mile refers to the piece of the network between the customer and the telco, CATV or wireless provider. Services are usually available in two classes of performance — guaranteed- and non-guaranteed-bandwidth. xDSL, wireless and cable modems are used to provide service in the latter category.
Non-guaranteed bandwidth services often make the last mile the least predictable part of the overall network, because broadband providers often have neighborhood concentrators or multiplexers that share a larger bandwidth among a number of users. Your expectation of available bandwidth may actually be based on your service provider's hypothetical calculation of each user's share of the system traffic at some point outside the peak traffic period. When you really need the bandwidth in the middle of your business day, it just might not be there.
A survival strategy is to contract for a larger share of the bandwidth than you would need if conditions were ideal. Remember that, if you are surfing a page with lots of images or a multimedia file while trying to use VoIP, the non-VoIP content may cause failure of the VoIP facility because of bandwidth congestion. By purchasing, say, double the bandwidth you theoretically need to support all your uses, including VoIP, you substantially increase your chances of having sufficient bandwidth even at peak traffic times.
For those without access to xDSL or other high-speed services, VoIP over dial-up V.90 may or may not work, depending upon a number of factors, especially your distance from the telco central office and the condition of the phone lines. VoIP requires a transmission path that provides a steady-state full-duplex throughput without interruptions or pauses.
Although more costly, guaranteed bandwidth services like ISDN or T1 are more likely to provide acceptable VoIP performance.
With VoIP over the public Internet, there are no guarantees of anything. One speech packet may travel to the destination using a direct FOTS connection, the next might go across the continent three times, and the third could end up arriving 300 ms late thanks to a satellite connection. The one after that might not arrive at all. For individuals and small organizations, however, this type of connection is often the only alternative available.
In the large corporate and government environments, it is much more likely that some form of WAN (wide area network) is available, providing guaranteed bandwidth between key locations. For business uses, it is prudent to avoid going "off-network" to reach smaller centers using VoIP technology. When the quality of your VoIP communications is unreliable, your customers, quite literally, "won't understand", and your corporate image may be adversely affected.
Some telcos provide a VoIP class of service with adequate quality at attractive rates; with other providers, the quality of service may be less than satisfactory. When considering choosing a third-party service provider, it pays to look closely at quality delivered and deliverable versus quality promised, especially if the rates are very low. As the old adage goes, when something seems too good to be true, it probably is.
IN THE FINAL ANALYSIS:
As with most products, the VoIP marketplace offers many different grades of VoIP equipment. Each provides different quality and error-recovery performance as the network performance changes.
Any product meriting consideration should include a statement of performance, based on Mean Opinion Score or MOS as measured in accordance with the ITU-T P.800 and related standards. Although there are automated systems using a technique known as Perceptual Speech Quality Measurement or PSQM to generate synthetic MOS values, there is no substitute for real tests using real people. Performance testing done in accordance with international standards should ideally be done by a neutral third party to guarantee objective and uniformly reliable results.
Before making a significant investment in a VoIP product, prudent purchasers should arrange for a VoIP "shoot-out", namely, have MOS tests done on a selection of possible products under identical conditions, as closely as possible replicating the anticipated actual-use conditions. These tests should include a network simulator to control introduction of known network impairments. Also. part of the tested mix should always include a benchmark G.711 sample. Remember that comparing tests done under disparate conditions will lead to erroneous conclusions.
Purchasers should look for a statement from the manufacturer or supplier that the MOS tests have been carried out with a number of stipulated transmission impairments present. Most VoIP equipment works well under ideal conditions, but you need to know how the equipment will perform in the imperfect real world!
Before making their decisions, purchasers need a good appreciation of the qualities of different VoIP products, how they will integrate with the purchaser's PC and, local and wide-area network environments, and above all, what quality can realistically be expected under the purchaser's actual-use operating conditions. Where in-house IT and telecommunications support staff may lack time or experience to do the necessary checking and comparisons, it is often cost-effective to get some outside support for the decision from an expert consultant.
Disclaimer and Copyright
© 2006 - 2012 Acleris