3.6 SPREAD-SPECTRUM CDMA FOR PCS AND PCN
3.6.3 A High-Density Voice PCS System
We consider here an S-CDMA high-density network of microcells for a voice application of PCS similar in principle to the IS-95 digital cellular standard but less complex and optimized for microcellular networks. There are two fundamental differences between the high-density network of microcells and the IS-95 digital cellular standard:
• Full-duplex operation is achieved by using time division duplexing (TDD) rather than the FDD dictated by the given cellular frequency band allo- cation.
• A 32-kbps voice compression is assumed rather than the lower-data-rate compression techniques for digital cellular, which require more digital sig- nal processing with more time delays.
Both these differences reduce cost and complexity. Using TDD is consis- tent with PCS bands which are generally not allocated as separate transmit and receive bands, as in the cellular bands. Using 32-kbps voice compression reduces processing delays and implementation costs.
In this analysis, we assume a network of evenly distributed microcell sites and randomly distributed mobile voice radios. We assume the network is time-synchronized so that all cell sites have a common time reference (obtained, e.g., from GPS receivers). Each active mobile unit is slave to the corresponding cell-site radios and therefore synchronized to the basic net- work clock. Of course, in practical systems, time errors will occur and will affect overall performance and capacity. Note that conventional TDMA sys- tems also require a similar network time synchronization. Because of this assumed network synchronization, all base stations transmit TDD packe- tized bursts followed by all the mobile units transmitting their packetized bursts.
Figure 3.6 illustrates cell sites simultaneously transmitting their packetized bursts. Note that during this time interval, each mobile unit receives the data burst from its assigned cell sites together with interference from other cell sites. Signals that are transmitted by their assigned cell site and are intended for other mobile units assigned to the same cell cite use orthogonal code- words.
During the inbound time frame, all mobile units transmit their packets to their cell sites. As shown in Figure 3.7, each cell-site receiver must handle reception of all the mobile unit signals assigned to its cell under interference from all mobile units in other cells.
In this analysis, we assume S-CDMA for both inbound and outbound channels. For microcells in which multipath time delays are smaller and
Figure 3.6. Outbound simulation model (base to handset).
Figure 3.7. Inbound simulation model (handset to base).
mobility is limited to roaming by pedestrians using these mobile units, this assumption is valid as long as all multipath delays are small compared to the chip-time interval.
Given ideal synchronization of cell sites, consider a direct-sequence spread-spectrum QPSK modulation in which we assume that interference derives from other similar synchronized radios. Here, we examine an exam- ple in which interference derives from cross-correlation stemming from the specific codewords used in the network. Propagation delays and time off- sets are taken into consideration in this simulation.
The system considered here is a spread-spectrum S-CDMA design with full-duplex operation using TDD. The following are its signal carrier fea- tures:
A PCS S-CDMA System
Type of modulation DS/QPSK
Voice transmission 32 kbps ADPCM digital voice Data transmission 4.8 kbps signaling channel
Processing gain 32 chips/bit
QPSK symbol burst rate 1.536 M symbols/sec
The S-CDMA system requires time synchronization for the network of microcells on the order of a burst chip-time interval, which is 651 nanosec- onds (nsec). This delay is much larger than the multipath delays expected for microcells, especially for indoor applications. The digital European cord- less telephone (DECT), a TDMA system to be discussed later, has a burst bit rate of 1.152 mbps, which requires time synchronization on the order of a bit time interval of 868 nsec. Thus, time synchronization issues are similar for these two systems.
3.6.3.1 Bit-Error Probabilities
Most textbook analyses of communication systems are based on the assumption that the receiver front-end noise is the limiting performance fac- tor. Even when performance limitations derive from other sources of inter- ference in the communication channel, most analyses approximate this interference as noise. For communication systems, the usual performance measure is bit error probability, with interference modeled as white Gaussian noise with spectral density denoted N0. The bit error probability, Pb, is typically expressed in terms of the ratio of the energy per bit,Eb, to noise,Eb/N0.
Except for a few cases, it is generally difficult to arrive at an exact closed- form expression for the bit error probability of a communication system, even under the assumption of white Gaussian noise interference.When inter- ference is not modeled as white Gaussian noise, analysis is even more diffi-
cult. For this reason, very few expressions for bit error probability are given for interference that is not white Gaussian noise.
For coherent BPSK, the bit error probability against white Gaussian noise is well known and is given by
(3.4) whereQ(x) is the Gaussian integral function
(3.5) For this ideal noise-limited case, we have the following required values of Eb/N0:
(3.6)
By using differentially coherent detection of DPBSK (assuming, again, that the limiting factor is white Gaussian noise), the bit error probability is given by the simple expression
(3.7) For this idealized case, the required Eb/N0is as follows:
(3.8)
Note that the difference between coherent detection and differentially coherent detection is less than 2 dB.
The above bit error probability expressions also apply to spread-spectrum communication systems of the direct-sequence type when the basic modu- lation is BPSK and the interference is white Gaussian noise. For our spread- spectrum system, however, the performance limitation is due to interference from other similar spread-spectrum radios.
In general, applications are characterized by a high density of users with mobile units roaming in an area with many microcells. This situation is ide- alized in Figure 3.8. In the S-CDMA system design, during the outbound frame, each mobile unit will be receiving signals from its assigned base sta- tion, as shown in Figure 3.6. It could also be receiving interference from neighboring base stations. For the inbound case, all the mobile units will be sending interfering signals to neighboring base stations. However, thanks to our use of orthogonal codes in the S-CDMA system, there is no interference among the radios within a microcell.
Eb
N0
⫽ à
5.92 dB, for Pb⫽10⫺2 7.93 dB, for Pb⫽10⫺3 9.30 dB, for Pb⫽10⫺4 10.34 dB, for Pb⫽10⫺5.
Pb⫽12e⫺Eb>N0.
Eb
N0
⫽ à
4.32 dB, for Pb⫽10⫺2 6.79 dB, for Pb⫽10⫺3 8.40 dB, for Pb⫽10⫺4 9.59 dB, for Pb⫽10⫺5. Q1x2⫽ 1
12p冮xqe⫺t2>2dt.
Pb⫽Q122Eb>N02
In Appendix 3B, we derive bounds on the bit error probability for our spread-spectrum system when interference derives from other similar spread-spectrum signals in a network of microcells. In general, obtaining exact expressions for the bit error probabilities is difficult, and we must resort to using easy-to-evaluate upper bounds on them even though upper bounds give worst case analyses, since the true error probabilities are less than the more easily computed upper bounds.
We show in Appendix 3B that in the interference-limited case, the bit error probability bound for direct-sequence BPSK signals is given by
(3.9) where SJR is the signal-to-interference ratio
(3.10) with
(3.11) HereCkis the cross-correlation of the interfering signal’s codeword with the receiver’s codeword, and Ais the signal amplitude. Note that this expres- sion is exactly the same as the bit error probability for differentially coher- ent detection of DBPSK signals if we replace N0byJ0/2.
J0⫽ a
k
Ck2. SJR⫽A2>J0
Pb 6 12e⫺SJR
Figure 3.8. Simulation model (random distribution).
The results in Appendix 3B are for the direct-sequence spread-spectrum coherent BPSK modulation with coherent modulation. The corresponding results for differentially coherent demodulation of DBPSK are expected to be within 2 dB of the bit error bounds given here. The DQPSK system may add an additional decibel of required energy per bit to interference. Using convolutional codes with Viterbi decoding will improve this performance by 4 to 6 dB.
3.6.3.2 Computer Simulations
A computer program has been developed to calculate the signal-to-inter- ference ratios for the idealized network of direct-sequence S-CDMA spread- spectrum radios. It allows various parameters to be changed to examine the effects of different codewords, power attenuation constants, voice activation factors, and numbers of randomly placed mobile handsets throughout the network. Naturally, the greater the power attenuation factor assumed, the lower the mutual interference between radios. A 35 percent voice activation level was also assumed.
Figure 3.9 shows a contour plot of the signal-to-interference ratio, SJR, for a mobile handset as a function of its location. As shown in Figure 3.9, the interference for mobile units derives from the cell sites of other micro- cells. The lack of symmetry is explained by the specific choice of codewords in each of the microcells in this model. Here, we assume six active voice chan- nels per microcell. Modified five-stage LFSR sequences were used.The prop- agation attenuation is assumed to be the inverse third power of distance.
Figure 3.9. Three-dimensional contour SJR plot for handset.
For the mobile-to-cell-site channel, the SJR at the cell-site receiver depends on the particular distribution of mobile handsets throughout the network. Figure 3.10 illustrates a typical example as a function of the num- ber of randomly distributed mobile handsets per microcell for the inbound direction.
3.6.3.3 Other System Issues
In the IS-95 digital cellular standard, mobile units synchronize on the cell site by locking onto the strongest pilot signal which is continuously trans- mitted by each of the cell sites. In this TDD structure, a similar broadcast pilot signal can be sent by each cell site at the beginning of its TDD burst.
In this TDD design, the cell-site packetized bursts are longer than the mobile units’ packetized bursts to accommodate the additional time needed to send a short common broadcast burst. This short burst serves as a pilot signal for the mobile units for maintaining a lock on the specific cell-site sig- nal. For initial acquisition, one of the cell-site signals is used for this purpose.
As in IS-95, this signal aids in the initial acquisition and provides paging information to the mobile units. Cell sites using different codewords for this purpose are used instead of offsets of the same PN sequence, as in the Qualcomm system. To achieve the rapid acquisition needed for TDD sys- tems, each receiver should use matched filters rather than the serial corre- lators typically used with longer PN sequences.
Soft handoff can be achieved by using a RAKE receiver processor as in the IS-95 system. Another alternative is to allow the receiver to track more
Figure 3.10. SJR versus number of users per cell.
than one signal with different codewords. With programmable matched fil- ters, this can be done in practice with very few additional gates in a custom chip design.
3.6.3.4 Comparison with DECT
For PCS applications, the current DECT system, using TDMA, serves as a baseline for comparison with the S-CDMA system described here. DECT has been under development in Europe for several years and represents the state of the art for PCS using conventional TDMA radios. These PCS sys- tems, which involve users carrying portable or mobile handsets while roam- ing in an area covered by many microcells, are designed for microcells of up to approximately 200 m in diameter. The most important criterion for com- parison is capacity, as measured by the total number of simultaneous voice users in a given high-density area (1 km2, for example), in which the num- ber of microcell sites and the total bandwidth within each area are constant.
The comparison should also assume that overall complexity and cost are roughly equivalent, although these can be difficult to measure.
As pointed out earlier, the design philosophy regarding spread-spectrum radios is different: Now interference is tolerated, and the primary issue is to design to a tolerable level of interference. This new design criterion for a net- work of microcells adds another design dimension that does not apply with conventional narrowband radios and places a much stronger emphasis on the quality of radios developed and the overall system design and signal coordination. Radios that tolerate greater interference afford more flexibility in designing a network of microcells, which is especially important since it facilitates the implementation of various antenna techniques at the base sta- tions, which can in turn dramatically increase microcell capacity.
DECT Standard
Within the 1.88—1.90 GHz band are 10 FDMA carriers. In each, 12 digital voice channels use TDMA for multiple-access and TDD for full-duplex oper- ation, given a total of 120 voice channels. The spacing between the FDMA carriers is 1.728 MHz.
Type of modulation GMSK
Voice transmission 32 kbps ADPCM digital voice Data transmission 4.8 kbps signaling channel Duplex voice channels per carrier 12
Time frame 10 msec
Time slot per channel 0.417 msec
(including guard space)
Number of time slots per frame 24
Burst bit rate 1.152 Mbps
Output power 250-mW burst peak
Note that with 24 time slots, the guard times needed between slots account for 23.3 percent of the total time frame. Each voice channel with TDD, over- head, and time gaps uses 96 kbps of the 1156-kbps burst rate.
This DECT design permits 12 voice users in every 1.728 MHz of band- width. By normalizing to 1 MHz, we have 6.94 active voice users per mega- hertz of bandwidth. In a large network of densely packed microcells, the available total bandwidth per cell will be reduced by a factor of because adjacent microcells cannot share the same frequency band. The frequency- of-reuse limitation of conventional narrowband radios can reduce overall capacity for the area covered by all microcells by as much as one-seventh.
Thus, this DECT system used in a high density of microcells averages only one active channel per megahertz per cell.
In the above S-CDMA example, we see that the bandwidth is roughly the same, but the potential number of active voice channels per cell is much larger for signal-to-interference ratios as high as 15 dB. This can be seen in the number of users allowed in the simulation results shown in Figures 3.9 and 3.10, in which 100 percent frequency of reuse is allowed.