Cellular and personal communications service (PCS) communications systems have historically been designed with voice traffic in mind. The patterns associated with voice communications are well known, having been observed since the invention and widespread use of the telephone. Voice can be characterized as relatively predictable, with each party talking about half the time in an interactive manner. The statistics of call duration and time of day are well understood, allowing traffic engineers to use a standard methodology to estimate the amount of capacity needed in a communications system. The wireline telephone network has been engineered in a hierarchical fashion using large circuit switches to efficiently connect one voice user to another. The physical circuit over which a call is made is held open for the entire duration of a call, hence the term circuit switching. Voice in wireline and mobile settings have similar characteristics. Existing cellular telephone systems have therefore been designed in a similar way and optimized to efficiently provide voice service.

Data traffic differs from voice in at least three important ways:

1. First, data traffic is much more unpredictable than voice traffic. Data is characterized as bursty, meaning that there is significant variability in when the traffic arrives, the rate at which it arrives, and the number of bits in the messages.

2. Second, data has very different requirements in terms of reliability. Whereas voice is very robust and capable of being understood even in a noisy environment with a high bit-error rate (BER) approaching one percent, data applications require extremely reliable delivery, with virtually no tolerance for bit errors. Because some bit errors are unavoidable on wireless links, it is important that fast and efficient recovery schemes are implemented to get the correct data bits to the application. A combination of forward error correction (FEC) and fast acknowledgements (an automatic retransmission request [ARQ]) satisfies this need. Powerful FEC is employed to dramatically reduce the BER, and an ARQ is used to guarantee reliable delivery.

3. Third, data traffic encompasses a much different and wider range of services than voice. Different types of services have different requirements along several dimensions. A data service can be characterized by its importance or priority. This is determined by the quality of service (QoS) that is required, which can be measured by the amount of delay that a user is willing to tolerate, and the reliability required. A service provider may offer differing service rates, or classes of service (CoS), accordingly. Premium service users may be given priority over best-effort users, whose traffic is sent if there is capacity available at the time. The data rate that is required and granted to a user is another dimension for a data service. A user may have a service-level agreement (SLA), which guarantees a certain minimum rate and allows a maximum or average rate over some period of time. A final aspect of a data service is latency, or response time. This determines the degree of interactivity that can be achieved, which is a measure of how quickly channel resources can be assigned at the request of a user.

Many feel that data services differ from voice in one other way, which is related to the variability in capacity demanded by the end user. If data users are allowed to consume as much bandwidth as they can, provided that there are no higher-priority users contending for resources, a system will tend to always be in a state of high utilization. The admission control mechanism, which governs how users access the system and are allocated resources, becomes a potential bottleneck under such circumstances. To provide low latencies in a wireless environment, where errors are unavoidable and packets must be retransmitted, it is necessary for a system to employ a fast ARQ capability so that packets received in error can be quickly retransmitted. Fast ARQ, in turn, requires that the user's access to the system be quick in order to transmit acknowledgements upon the receipt of correct packets. Systems employing a contention-based admission control generally exhibit ever-growing latencies as utilization increases and cannot support a fast ARQ capability. A well-designed system employing scheduled, non-contention-based access can yield much lower latency and support rapid ARQs.

Today's data traffic is primarily driven by wireline users and generated by a broad range of sources and applications, including Internet use, electronic messaging, file transfer, and, to some degree, voice traffic such as voice over Internet protocol (VoIP) (using primarily desktop and portable computers). The transmission control protocol (TCP)/Internet protocol (IP) suite, which is the data-transmission protocol used for the Internet, is the most widely used and governs the bulk of all data traffic.

In the future, the number of devices generating data traffic is expected to skyrocket by an order of magnitude. Information appliances and other data-centric wireless and consumer products will fuel this growth, leveraging the inherent value that mobility provides.

Figure 1. Wireless User versus Data-Device Growth
Sources: Cahners In-Stat, Goldman Sachs, IDC, Lehman Brothers, Semico

The types and mix of applications and services used will also change over time as the mix of devices changes, with more interactive applications such as voice, gaming, two-way messaging, and even video joining less-interactive applications such as streaming audio and video and traditional Web browsing. Much of this traffic will go over wireless data networks.

The true potential for ubiquitous wireless data communications can be unleashed when end-user devices efficiently support native TCP/IP connectivity, without the need for special translators and filters. Mobile wireless communications has traditionally posed a difficult performance challenge for TCP/IP protocols. TCP was designed and optimized around reliable wireline links, where BERs and packet-error rates are substantially lower than that typically achievable via wireless. When TCP encounters dropped or lost packets, it assumes that there is congestion on the link, but not that the link itself is unreliable. Congestion is handled by reducing the information rate at which the sender is allowed to transmit. By interpreting the unavoidable errors that occur in a wireless environment as congestion, the effective data rate seen by the end user is reduced. This is further compounded by the fact that the initial data rate at the start of a TCP session is low—far below the ultimate peak rate—and gradually builds over time as the systems figure out where the peak rate is. This slow-start aspect of TCP can dramatically add to latencies as the link is throttled down due to errors and then slowly ramped back up.

In summary, there is a much wider range of requirements and characteristics for data communications than there is for voice. This variability prohibits data from being efficiently carried over the hierarchical networks designed for voice traffic, whether wireline or wireless. Mobile data systems face additional challenges as a result of the vagaries of the wireless environment.