Part IV MULTIPLE ACCESS AND ADVANCED TRANSCEIVER SCHEMES 363
18.3 Cellular Code-Division-Multiple-Access Systems
18.3.1 Principle Behind Code Division Multiple Access – Revisited
When analyzing the multiaccess capability of a system, we are essentially asking the question “What prevents us from serving an infinite number of users at the same time?”. In a TDMA/FDMA system, the answer is the limited number of available timeslots/frequencies. Users can occupy those slots, and not interfere with each other. But when all possible timeslots have been assigned to users, there are no longer free resources available, and no further users can be accepted into the system.
In a CDMA system, this mechanism is subtly different. We first analyze the uplink in the following, where spreading codes are imperfect, but there is a large number of them. Different users
are distinguished by different spreading codes; however, as user separation is not perfect, each user in the cell contributes interference to all other users. Thus, as the number of users increases, the interference for each user increases as well. Consequently, transmission quality decreases gradually (graceful degradation), until users find the quality too bad to place (or continue) calls. Consequently, CDMA puts a soft limit on the number of users, not a hard limit like TDMA. Therefore, the number of users in a system depends critically on the Signal-to-Interference-and-Noise Ratio (SINR) required by the receiver. It also implies that any increase in SINR at the receiver, or reduction in the required SINR, can be immediately translated into higher capacity.
Most interference stems from within the same cell as the desired user, and is thus termedintracell interference. Total intracell interference is the sum of many independent contributions, and thus behaves approximately like Gaussian noise. Therefore, it causes effects that are similar to thermal noise. It is often described by noise rise– i.e., the increase in “effective” noise power (sum of noise and interference power) compared with the noise alone(N0+I0)/N0. Figure 18.8 shows an example of noise rise as a function of system load; heresystem load is defined as the number of active users, compared with the maximum possible number MC. We see that noise rise becomes very strong as system load approaches 100%. A cell is thus often judged to be “full” if noise rise is 6 dB. However, as mentioned above, there is no hard limit to the number of active users.
20 18 16 14 12 10 8 6 4 2 0
Noise rise (dB)
6 dB
10 20 30 40 50 60 70 80 90 100
System load h (%)
75%
0
Figure 18.8 Noise rise as a function of system load in a code-division-multiple-access system.
Reproduced with permission from Neubauer et al. [2001]©T. Neubauer.
Some interference is from neighboring cells, and thus calledintercell interference. A key property of a CDMA system is that it usesuniversal frequency reuse(also known asreuse distance one). In other words, the same frequency band is used in all cells; users in different cells are distinguished only by different codes. As discussed in Section 18.2.6.1, the amount of interference is mostly determined by the codes that are used in the different cells.
Spread Spectrum Systems 403
Many of the advantages of CDMA are related to the fact that interference behaves almost like noise, especially in the uplink. This noise-like behavior is due to several reasons:
• The number of users (and therefore, of interferers) in each cell is large.
• Power control makes sure that all intracell signals arriving at the BS have approximately the same strength (see also below).
• Interference from neighboring cells also comes from a large number of users. Spreading codes are designed in such a way that all signals in one cell have approximately the same cross-correlation with each signal in all neighboring cells. Note that this implies that we cannot simply reuse the same codeset in each cell; otherwise, there would be one user in the neighboring cell that would contribute much more interference (the user that uses the same code as the desired user in the desired cell).
Due to the above effects, total interference power shows very little fluctuations. At the same time, the power control makes sure that the signal strength from the desired user is always constant.
The SINR is thus constant, and no fading margin has to be used in the link budget. However, note that making interference as Gaussian as possible is not always the best strategy for maximizing data throughput; multiuser detection (Section 18.4) actively exploits structure in interference and works best when there are only a few strong interferers.
In the downlink, spreading codes are orthogonal, so that (at least in theory) different users can be separated completely. However, in this case, the number of users in the cell is limited by the num- ber of Walsh–Hadamard codes. The situation then becomes similar to a TDMA system: if theMC available Walsh–Hadamard codes are used up, then no further users can be served. Furthermore, there is also interference from the Walsh–Hadamard codes of neighboring cells. The situation can be improved by multiplying the Walsh–Hadamard codes by ascrambling code. Walsh–Hadamard codes that are multiplied by the same scrambling code remain orthogonal; codes that are multi- plied by different scrambling codes do not interfere catastrophically. Therefore, different cells use different scrambling codes (see also Chapter 26).
Downlink intercell interference doesnot come from a large number of independent sources. All the interference comes from the BSs in the vicinity of the considered MS – i.e., a few (at most six) BSs constitute the dominant source of interference. The fact that each of these BSs transmits signals to a large number of users within their cell does not alter this fact: the interfering signal still comes from a single geographical source that has a single propagation channel to the victim MS. Consequently, the downlink might require a fading margin in its link budget.
18.3.2 Power Control
As we mentioned above, power control is important to make sure that the desired user has a time-invariant signal strength, and that the interference from other users becomes noise-like. For further considerations, we have to distinguish between power control for the uplink and that for the downlink:
• Power control in the uplink: for the uplink, power control is vital for the proper operation of CDMA. Power control is done by a closed control loop: the MS first sends with a certain power, the BS then tells the MS whether the power was too high or too low, and the MS adjusts its power accordingly. The bandwidth of the control loop has to be chosen so that it can compensate for small-scale fading – i.e., has to be on the order of the Doppler frequency. Due to time variations of the channel and noise in the channel estimate, there is a remaining variance in the powers
arriving at the BS; this variance is typically on the order of 1.5–2.5 dB, while the dynamic range that has to be compensated is 60 dB or more. This variance leads to a reduction in the capacity of a CDMA cellular system of up to 20% compared with the case when there is ideal power control.
Note that an open control loop (where the MS adjusts its transmit power based on its own channel estimate) cannot be used to compensate for small-scale fading in a Frequency Domain Duplexing (FDD) system: the channel seen by the MS (when it receives signals from the BS) is different from the channel it transmits to (see Section 17.5). However, an open loop can be used in conjunction with a closed loop. The open loop compensates for large-scale variations in the channel (path loss and shadowing), which are approximately the same at uplink and downlink frequencies. The closed loop is then used to compensate for small-scale variations.
• Power control in the downlink: for the downlink, power control is not necessary for CDMA to function: all signals from the BS arrive at one MS with the same power (the channel is the same for all signals). However, it can be advantageous to still use power control in order to keep the total transmit power low. Decreasing the transmit power for all users within a cell by the same amount leaves unchanged the ratio of desired signal power to intracell interference – i.e., interference from signals destined for other users in the cell. However, it does decrease the power of total interference to other cells. On the other hand, we cannot decrease signal power arbitrarily, as the SNR must not fall below a threshold. The goal of downlink power control is thus to minimize the total transmit power while keeping the BER or SINR level above a given threshold. The accuracy of downlink power control need not be as high as for the uplink; for many cases, open loop control is sufficient.
It is worth remembering that the power control of users in adjacent cells does not give constant power of the intercell interference. A user in an adjacent cell is power controlled by its own BSs – in other words, power is adjusted in such a way that the signal arriving at its desired BS is constant. However, it “sees” a completely different channel to the undesired BS, with temporal fluctuations that the desired BS neither knows nor cares about. Consequently, intercell interference is temporally variant.
Interference power from all users is the same only if all users employ the same data rate.
Users with higher data rates contribute more interference power; high-data-rate users can thus be a dominant source of interference. This fact can be understood most easily when we increase the data rate of a user by assigning multiple spreading codes to him. In this case, it is obvious that the interference this user contributes increases linearly with the data rate. While this situation did not occur for second-generation cellular systems, which had only speech users, it is certainly relevant for third-generation cellular systems, which foresee high-data-rate services.
It should also be noted that power control is not an exclusive property of CDMA systems; it can also be used for FDMA or TDMA systems, where it decreases intercell interference and thus improves capacity. The major difference is that power control isnecessaryfor CDMA, while it is optional for TMDA/FDMA.
Soft Handover
As all cells use the same frequencies, an MS can have contact with two BSs at the same time. If an MS is close to a cell boundary, it receives signals from two or more BSs (see Figure 18.9) and also transmits to all of these BSs. Signals coming from different MSs have different delays, but this can be compensated by the Rake receiver, and signals from different cells can be added coherently.7
7Note that different cells might use different codes. This is not a major problem; it just means that (for the downlink) different correlators in the fingers of the Rake receiver have to use different spreading sequences.
Spread Spectrum Systems 405
BSs and MS use the same code sequence
BS BS BS
MS
Figure 18.9 Principle behind soft handover.
Reproduced with permission from Oehrvik [1994]©Ericsson AB.
This is in contrast to the hard handover in an FDMA-based system, where an MS can have contact with only one BS at a time, because it can communicate only on one frequency at a time.
Consider now an MS that starts in cell A, but has already established a link to BS B as well.
At the outset, the MS gets the strongest signal from BS A. As it starts to move toward cell B, the signal from BS A becomes weaker, and the signal from BS B becomes stronger, until the system decides to drop the link to BS A. Soft handover dramatically improves performance while the MS is near the borderline of the two cells, as it provides diversity (macrodiversity) that can combat large-scale as well as small-scale fading. On the downside, soft handover decreases the available capacity in the downlink: one MS requires resources (Walsh–Hadamard codes) in two cells at the same time, while the user talks – and pays – only once. Furthermore, soft handover increases the amount of signaling that is required between BSs.
18.3.3 Methods for Capacity Increases
• Quiet periods during speech transmission: for speech transmission, CDMA makes implicit use of the fact that a person does not talk continuously, but rather only about 50% of the time, the remainder of the time (s)he listens to the other participant. In addition, there are pauses between words and even syllables, so that the ratio of “talk time” to “total time of a call” is about 0.4.
During quiet periods, no signal, or a signal with a very low data rate, has to be transmitted.8In a CDMA system, not transmitting information leads to a decrease in total transmitted power, and thus interference in the system. But we have already seen above that decreasing the interference power allows additional users to place calls. Of course, there can be a worst case scenario where all users in a cell are talking simultaneously, but, statistically speaking, this is highly improbable, especially when the number of users is large. Thus, pauses in the conversation can be used very efficiently by CDMA in order to improve capacity (compare also discontinuous transmission in TDMA systems).
• Flexible data rate: in an FDMA (TDMA) system, a user can occupy either one frequency (times- lot), or integer multiples thereof. In a CDMA system, arbitrary data rates can be transmitted by
8Actually, most systems transmitcomfort noiseduring this time – i.e., some background noise. People speaking into a telephone feel uncomfortable (think the connection has been interrupted) if they cannot hear any sound while they talk.
an appropriate choice of spreading sequences. This is not important for speech communications, which operate at a fixed data rate. For data transmission, however, the flexible data rate allows for better exploitation of the available spectrum.
• Soft capacity: the capacity of a CDMA system can vary from cell to cell. If a given cell adds more users, it increases interference to other cells. It is thus possible to have some cells with high capacity, and some with lower; furthermore, this can change dynamically, as traffic changes.
This concept is known asbreathing cells.
• Error correction coding: the drawback of error correction coding is that the data rate that is to be transmitted is increased, which decreases spectral efficiency. On the other hand, CDMA con- sciously increases the amount of data to be transmitted. It is thus possible to include error correction coding without decreasing spectral efficiency; in other words, different users are distinguished by different error correction codes (coding by spreading). Note, however, that commercial systems (UMTS, Chapter 26) do not use this approach; they have separate error correction and spreading.
18.3.4 Combination with Other Multiaccess Methods
CDMA has advantages compared with TDMA and FDMA, especially with respect to flexibility, while it also has some drawbacks, like complexity. It is thus obvious to combine CDMA with other multiaccess methods in order to obtain the “best of both worlds.” The most popular solution is a combination of CDMA with FDMA: the total available bandwidth is divided into multiple subbands, in each of which CDMA is used as the multiaccess method. Clearly, frequency diversity in such a system is lower than for the case where spreading is done over the whole bandwidth.
On the positive side, the processing speed of the transmitters and receivers can be lower, as the chip rate is lower. The approach is used, e.g., in IS-95 (which uses 1.25-MHz-wide bands) and Universal Mobile Telecommunications System (UMTS), which uses 5-MHz-wide subbands.
Another combination is CDMA with TDMA. Each user can be assigned one timeslot (as in a TDMA system), while the users in different cells are distinguished by different spreading codes (instead of different frequencies). Another possibility is to combine several timeslots, and build up a narrowband CDMA system within them. This system works best when adding a CDMA component to an existing TDMA system (e.g., the TDD mode of UMTS).