Code Division Multiple Access

CDMA systems employ wide-band signals with good cross-correlation properties [Kohno, Meidan, and Milstein, 1995]. That means, the output of a filter matched to one user’s signal is small when a different user’s signal is input. A large body of work exists on spreading sequences which lead to signal sets with small cross-correlations [Sarwate and Pursley, 1980]. Because of their noise-like appearance such sequences are often referred to as pseudo-noise (PN) sequences and because of their wide-band nature CDMA systems are often called spread-spectrum systems.

Spectrum spreading can be achieved mainly in two ways: through frequency hopping as explained above or through direct sequence spreading. In direct sequence spread spectrum a high-rate, antipodal pseudo-random spreading sequence modulates the transmitted signal such that the bandwidth of the resulting signal is roughly equal to the rate of the spreading sequence. The cross-correlation of the signals is then largely determined by the cross-correlation properties of the spreading signals. Clearly, CDMA signals overlap in both time and frequency domain but are separable based on their spreading waveforms.

An immediate consequence of this observation is that CDMA systems do not require tight synchronization like TDMA systems. By the same token, frequency planning and management are not required as frequencies are re-used throughout the coverage area.

Propagation Considerations

Spread spectrum is well suited for wireless communication systems because of its “built-in” frequency diversity. As discussed before, in cellular systems the delay spread measures several microseconds and, hence, the coherence bandwidth of the channel is smaller than 1 MHz. Spreading rates can be chosen to exceed the coherence bandwidth such that the channel becomes frequency selective, i.e., different spectral components are affect in unequally by the channel and only parts of the signal are affected by fades. Expressing the same observation in time domain terms, multi-path components are resolvable at a resolution equal to the chip period and can be combined coherently for example by means of a RAKE receiver [Proakis, 1989]. An estimate of the channel impulse response is required for the coherent combination of multi-path components. This estimate can be gained from a training sequence or by means of a so-called “pilot” signal.

Even for cordless telephone systems, operating in environments with sub-microsecond delay spread and corresponding coherence bandwidths of a few MHz, the spreading rate can be chosen large enough to facilitate multi-path diversity. If the combination of multi-path components described above is deemed to complex a simpler, but less powerful, form of diversity can be used which decorrelates only the strongest received multi-path component and relies on the suppression of other path components by the matched filter.

Multiple-Access Interference

If it is possible to control the relative timing of the transmitted signals, like on the down-link, the transmitted signals can be made perfectly orthogonal and, if the channel only adds white Gaussian noise, matched filter receivers are optimal for extracting a signal from the superposition of waveforms. If the channel is dispersive because of multi-path the signals arriving at the receiver will be no longer orthogonal and will introduce some multiple-access interference, i.e., signal components from other signals which are not rejected by the matched filter.

On the up-link extremely tight synchronization to within a fraction of a chip period, which is defined as the inverse of the spreading rate, is generally not possible and measures to control the impact of multiple-access interference must be taken. Otherwise, the near-far problem, i.e., the problem of very strong undesired users’ signals overwhelming the weaker signal of the desired user, can severely decrease performance. Two approaches are proposed to overcome the near-far problem: power control with soft handovers and multi-user detection.

Power control attempts to ensure that signals from all mobiles in a cell arrive at the base station with approximately equal power levels. To be effective power control must be accurate to within about 1 dB and fast enough to compensate for channel fading. For a mobile moving at 55 mph and transmitting at 1 GHz, the Doppler bandwidth is approximately 100 Hz. Hence, the channel changes its characteristic drastically about 100 times per second and on the order of 1000 bit/s must be sent from base station to mobile for power control purposes. As different mobiles may be subject to vastly different fading and shadowing conditions a large dynamic range of about 80 dB must be covered by power control. Notice, that power control on the down-link is really only necessary for mobiles which are about equidistant from two base stations, and even then neither the update rate nor the dynamic range of the up-link is required.

The interference problem that arises at the cell boundaries where mobiles are within range of two or more base stations can be turned into an advantage through the idea of soft handover. On the down-link, all base stations within range can transmit to the mobile which in turn can combine the received signals to achieve some gain from the antenna diversity. On the up-link a similar effect can be obtained by selecting the strongest received signal from all base stations which received a user’s signal. The base station which receives the strongest signal will also issue power control commands to minimize the transmit power of the mobile. Note, however, that soft handover requires fairly tight synchronization between base stations and one of the advantages of CDMA over TDMA is lost.

Multi-user detection is still an emerging technique. It is probably best used in conjunction with power control. The fundamental idea behind this technique is to model multiple-access interference explicitly and devise receivers which reject or cancel the undesired signals. A variety of techniques have been proposed ranging from optimum maximum-likelihood sequence estimation via multi-stage schemes, reminiscent of decision feedback algorithms, to linear decorrelating receivers. An excellent survey of the theory and practice of multi-user detection was given by Verdu [Verdu, 1992].

Further Remarks

CDMA systems work well in conjunction with frequency division duplexing. This arrangement decouples the power control problem on the up-link and down-link, respectively.

Signal quality enhancing methods like time diversity through coding and interleaving can be applied just like with the other access methods. In spread spectrum systems, however, coding can be “built” in to the spreading process avoiding the loss of bandwidth associated with error protection. Additionally, CDMA lends itself naturally to the exploitation of speech pauses which make up more than half the time of a connection. If no signals are transmitted during such pauses then the instantaneous interference level is reduced and the total number of users supportable by the system can be approximately doubled.