Andrew Wright, Director, Product Research Oliver Nesper, DSP Design Engineer

1. Construction and Operation

Figure 1 below illuminates the commercial requirements for a Waveshaper product. The graph illustrates that the cumulative distribution function of the raw or intrinsic waveform is required to be modified such that the cumulative probabilty density function provided is met. The key feature is that waveshaping process must ensure that a particular amplitude threshold is not exceeded. This can be seen by the vertical descent of the second distribution function, which indicates that for this particular scenario amplitude excursions that exhibit crests greater than 5 dB above the average power do not occur.



Figure 1 Signal Statistics

Figure 2 illustrates the construction of the PALADIN Waveshaper kernel. Four individual WCDMA signal streams are accepted as the chip’s input. Each of them enters a base band preconditioning soft clipping stage, which is followed by a programmable pulse shaping and upsample-by-two filter stage. The signals are then further up-sampled by 4 in two half-band filter stages and followed by a modulation stage which frequency-converts the signals to individually specified carrier frequencies within a 20 MHz WCDMA frequency allocation. This chip supports a 1 Hz raster which offers significantly more precision than the required 200 kHz specified raster.



Figure 2 Waveshaper Kernel

All four up-sampled signal streams are then digitally combined and delayed prior to transmission. The delay stage is critical to successful operation for it permits the Predictive Decresting Waveform Generator to examine the entire waveform construction process, that is raw input data, preconditioned data, pulse shaped and frequency shifted data stream and to assess the probability and magnitude of a potential signal crest. This permits a waveform to be constructed and combined with the transmission stream that destructively interferes with the transmission signal’s crest to reduce the signal crest to below the predetermined customer-set threshold. This process is illustrated in Figure 3. A key and important property of the corrective waveform is that, should specific carrier allocations not be utilized, injection of energy into these allocations is not permissible. Figure 4 illustrates this important frequency domain characteristic. An additional and important property of the Predictive Decresting Waveform Generator is that it examines the composite waveform and individual component carrier power levels and manipulates the properties of the corrective waveform so that signal quality metrics for each individual carrier are equally modified. That is EVM, ACLR and Roe measurements for all channels will be equally impacted. The peak controlled and combined four-carrier signal is then up-sampled again to the final output rate and followed by a final frequency translation or DQM stage that can be by passed. After a sin(x)/x compensation stage the signal passes through the final clipping block that clips extremely rare peak events.



Figure 3 Basic Waveform Construction Process – Time Domain Analysis



Figure 4 Waveform Construction Process – Frequency Domain Analysis

2. Performance Results

This section will present some selected results that can be achieved using Waveshaper’s technology. The results are derived by considering the 10-4 probability point of peak occurrence and varying the desired maximum signal peak level. The results compare simulations for pure base band clipping and waveform compensation only, operating on 4-carrier TM1 WCDMA signals with 64 active DPCHs plus control channels. Results are shown for the 3GPP requirements that were identified in Section 2.1.

Figure 5 shows results for the resulting PAR versus EVM and the large gap between using base band clipping and pulse compensation is evident. Consider for example the 12% EVM point, where the PAR for base band clipping is around 9.2 dB and for pulse compensation around 7.1 dB, an improvement of more than 2 dB. Uncompensated signal PARs were around 10 dB.



Figure 5 Waveshaping vs. Baseband Clipping, PAR versus EVM

Results for the PCDE are shown in Figure 6. For the selected target of 12% EVM around -38dB PCDE is attainable for baseband clipping and pulse compensation. Again, the superior performance of the pulse compensation method is evident.



Figure 6 Waveshaping vs. Baseband Clipping, PAR versus PCDE

Considering the ALCR1  (Figure 7) of the processed signal, we can see the trade-offs that need to be made when using the pulse compensation. Base band clipping, as was noted before, is performed on the baseband signals before the pulse shaping filtering is applied. The amount of clipping applied has, therefore, no effect on the adjacent channel power leakage ratio. Using the pulse compensation we note that the method is not ideal as a certain amount of power leaks into adjacent bands. The limit set forth in the 3GPP specification is 45 dB and for our chosen operating point of 12% EVM we end up with 71dB ACLR1. Note that the compensation pulse shape in Waveshaper is programmable, which would permit trading off better ACLR1 performance for other signal measurements, if desired.




Figure 7 Waveshaping vs. Baseband Clipping, PAR versus ACLR1

The results we presented were generated using TM1 with 64 users and four carriers. Results for other test signals with a different numbers of carriers and different OVSF codes showed different PAR levels, but an overall similar characteristic between baseband and pulse compensation and an overall consistent performance gap as well.

Table 1 compares the improvement in signal PAR obtainable for the different compensation methods. As a reference point the uncompensated case is shown as well. We can see from these results that, depending on the parameters we can allow to trade-off, improvements of almost 3 dB can be obtained with leaving enough margin to the specification. This translates into significant savings in the base station’s power amplifier.

Table 1 Summary of Base Band Clipping versus Pulse Compensation Performance



About the Author

Andrew Wright is Director of Wireless and Signal Processing Product Research at PMC-Sierra. Dr Wright is a former co-founder and CTO of Datum Telegraphic Inc. and holds a Ph.D. in Microwave Engineering (meteorology). Since 1995, he has specialized in signal processing solutions for third generation wireless systems.

Oliver Nesper is a DSP Design Engineer in the Access Product Division. He has worked on the development of the PALADIN (Predistortion) and PALADIN Waveshaper products. Prior to that he was with Spectrum Signal Processing as a hardware development engineer working on the design of Soft Radio Receiver Platforms.

注解：1  ACLR2 was not affected by the signal processing performed.