How filters separate users in 3G

How filters separate users in 3G
The Universal Mobile Telecommunications System is a code-division multiple-access standard that contains two built-in standards: frequency-division duplex (FDD) and time-division duplex (TDD).

In the case of FDD, the basestation transmit frequency and the mobile transmit frequencies are widely separated. Because of this, a handset may interfere with the signal of another handset, but never with a basestation signal. A long code is employed to randomize the transmitted signals. In TDD, the basestation and handset share the same frequency, and there is no long code.

Both standards employ CDMA to replace a transmitted symbol by an orthogonal short-code sequence. As a result, the bandwidth of the signal at 5 MHz is much wider than in a second-generation GSM system. Also, signal-correlation operations-that is, the ability to correlate signals with users-allow multiple users to transmit in the same frequency band without interference. Because of this, the amount of energy in the signal due to one user is small. But this technique requires the channel filter to have a much higher sampling rate and a larger resolution than those of a single-channel filter for a GSM signal.

Digital filtering to remove out-of-band signals has the same purpose in any wireless communications system that employs a frequency division of signals. The driving requirement for the out-of-band rejection is the near/far problem. The near/far problem occurs when the signal band of interest originates from a distant source (a user who is remote from a basestation), and the interfering signal band next to this signal of interest is much higher in power (perhaps because it is closer).

Where channels are spaced closely together, larger-amplitude signals are more significant interferers than weaker ones. A large interfering signal will therefore require significant rejection to ensure it doesn't leak into the signal band of interest. For the FDD system, the power leakage into adjacent channels must be 45 dB down from the in-band power. For TDD, the requirement is only 33 dB.

In the case of UMTS the signal has a 4.68-MHz signal bandwidth, and the signals are spaced by 5 MHz, allowing a gap of 0.32 MHz between signals. This seems like a waste of bandwidth, but a closer spacing would make it harder to achieve the out-of-band rejection at the next carrier.

Transmission dependence

The quality of the transmitted signal and the transmit-filtering process are actually dependent on the signal being transmitted. This can be seen in the change in the UMTS FDD specification as the standard moved from Release 4 to Release 5. For instance, the error vector magnitude of the downlink transmitted signal-the amount of error tolerated-was reduced from 17.5 percent to 12.5 percent when the signal constellation moved from the four-phase quadrature phase-shift keying (QPSK) to the 16-constellation quadrature-amplitude modulation scheme (16QAM), which has a higher symbol rate but is more difficult to execute.

Other common communication system techniques are used in UMTS. The spectral shape is a raised cosine with 22 percent excess bandwidth. Squared-root raised-cosine filtering is used on each end. To achieve a subsample at the receiver, pre-emphasis is used to cancel the low-pass effect of the D/A converter.

Interpolation is also used at the receiver for all possible subsamples at once, because the "fingers" of the CDMA signal can appear as multiple subsample phases. For UMTS, four- to eight-times oversampling is used-but this filter is not considered a significant part of the Mips requirement for the baseband processor (that is, it may require separate hardware).

In practice, it may not be feasible to remove all out-of-band or in-band interference. In these cases, higher layers in the standard are developed to deal with interference using retransmission, which is already a deployed part of the GSM standard, or joint detection, which is being actively discussed for GSM. In GSM, retransmission is used along with frequency hopping to allow a much higher frequency reuse in neighboring cells.

Retransmission is the ability to send an acknowledgement to a packet so that the transmitter knows quickly whether it has to retransmit that packet because of failed reception. Joint detection is a form of active cancellation of in-band interferers because passive filtering is not possible. The interfering signal is detected and canceled. A similar technique used for CDMA is multiuser detection (MUD), which is a hot topic in academia but has yet to make a significant impact in UMTS systems.

For CDMA-based systems such as UMTS, the spreading code randomizes signals and provides spreading gain. For FDD systems, this means that even when the frequency reuse is 100 percent, there is no need to provide any kind of retransmission policy to support voice capacity, though in newer standard releases fast retransmission has been used to enhance the capacity for data traffic. In TDD systems, the possibility that a basestation transmission to a user in its cell could be drowned out by the transmission of a user in an adjoining cell means that retransmission is required and several forms of MUD are often proposed.

Some forms of filtering take place at the basestation amplifier. An increasingly used technique is power amplifier predistortion. Simple power amplification systems are too nonlinear to be used in a UMTS system. The traditional feedback linearization techniques lead to too much amplitude loss.

Until recently, feed-forward techniques were used. In these amplifiers, samples of the distortion produced by the power amplifier are amplified and subtracted from the output of the PA. This requires delicate analog delays and a significant increase in the number of power transistors.

Predistortion for PAs has become a popular academic topic and is starting to appear in real systems. The basic predistortion system demodulates a sample of the power amplifier output and compares it to the input signal. Complex signal processing on a DSP is used to calculate the values of a nonlinear predistortion filter that processes the signal before it enters the power amplifier.

This predistorter is a memoryless, polynomial nonlinearity. In more complex predistortion, variants of Volterra filters are used. These filters are very wideband because the nonlinearity causes spectral growth. A 5x spectral growth is not uncommon. That necessitates a filter bandwidth requirement of 25 MHz, leading to implications for sampling rate and resolution.

In an FDD system, equalization is not common. The energy in the wideband signal is separated out using fingers that separately demodulate the signal. The finger outputs are combined using a maximal-ratio combiner (MRC), which is a maximum-likelihood linear receiver if the channels for each finger are independent. For smaller spreading factors, the Mips requirement for MRC can be significant compared with the rest of the demodulation process, except in the case of voice rates.

In TDD systems, the lack of a long code and lower spreading factor increases the correlation between signals. Equalization is used for multiple signals being jointly decoded at the basestation. But equalization is more complex than MRC. The channel estimation function is also challenging, since it involves matrix manipulation on matrices with dimensions equal to the number of fingers.

See related chart

http://www.eet.com