Digital predistortion is an approach to amplifier linearization that permits the efficiency of the multi-carrier amplifier to be dramatically increased. The principle of predistortion is intrinsically very simple: a non-linear distortion function is built in the numerical digital baseband signal processing domain that is commensurate ("equal") but opposite to the distortion function exhibited by the amplifier. A highly linear, distortion-free system is achieved when these two non-linear distortion functions are combined.
The beauty of this approach is that the analog power amplifier is now permitted to become a simple class-AB platform, freeing vendors from the burden and complexity of manufacturing complex feed-forward amplifiers. Moreover, since the amplifier no longer needs the error amplifier distortion correction circuitry, system efficiency is significantly enhanced.
The objective of a digital predistortion system is to numer ically generate in the real-time digital complex baseband signal processing domain a non-linearity that has a complimentary characteristic to that exhibited by the amplifier. If the baseband non-linearity is correctly constructed then the overall system response to a signal that flows serially through the cascade of the baseband non-linearity and the amplifier is that of a linear gain response. This linear gain response is highly desirable since it implies that distortion and spectral regrowth will not occur.
Ordinarily, radio engineers are predominantly concerned with both AM-AM and AM-PM distortion. These distortion mechanisms are referred to as memoryless and correspond to the belief that the instantaneous distortion observed at the output of the amplifier can be directly mapped to the instantaneous amplitude of the signal driving the amplifier input. This distortion mechanism represents the bulk of the amplifier distortion characteristic.
However, eliminating this bulk di stortion mechanism is not sufficient to entirely eliminate all spectral regrowth generated by the amplifier because small, residual non-linear memory effects are present.
The exact definition of a non-linear memory effect is often subject to debate. But a practical working definition is that the current and the previous input stimuli affect the current output of the amplifier. Furthermore, this relationship between the current output and the current and previous input stimuli is not restricted to being linear. In practice, power amplifiers exhibit several distinct non-linear memory characteristics, which are distinguished by substitutionally different time constants.
Following the basic principles of predistortion, if a linear system is to be constructed from a cascade of non-linearities and the amplifier is classified as a weak Volterra kernel, then the complimentary non-linearity will also require the construction of a Volterra kernel. This is the essence of the PMC-Sierra Paladin predistor tion product family.
The topic of amplifier efficiency often leads to much consternation when examined for the first time. Referring to amplifier textbooks will often result in discussions comparing the merits of class A, class AB, class B and class C amplifiers in terms of gain, power utilization factor and efficiency. It is not uncommon for these to state that the theoretical efficiencies of a Class A amplifier is 50%, which rises to 70% as the conduction angle is reduced and class AB operation is invoked. Class B and Class C amplifiers offer efficiencies that theoretically exceed 70% but with severe non-linearity and diminishing gain. The efficiency numbers quoted are often at odds with the power added efficiencies of 5% to 20% that are observed in practice.
Further investigation readily resolves this conundrum. The key issue to realize is that theoretical efficiencies are based upon the assumption that the amplifier will amplify a RF sinusoid whose peak-to-peak variation exercises the e ntire load line of the amplifier an active transistor or FET from cut-off/turn-on to full power saturation.
It is under these circumstances that Class A amplifiers yield efficiencies of 50%, while class AB amplifiers yield efficiencies of 70%. The lost energy is utilized to support the quiescent bias operating condition of the amplifier. These efficiency benchmarks rapidly degrade when information-bearing signals are amplified because the operating point of the amplifier and input drive levels are set up to ensure that signal peaks just exercise the saturation or maximum output power point of the amplifier.
When these signal peaks or crests occur, the amplifier does approach its theoretical operating efficiencies since the waveform at RF does appear as a large amplitude sinusoid for a very short duration of time. However, the majority of the time, the average operating point excursion, defined by some stochastic mean or "average " excursion, is substantially less than that of the peak handling capability of the amplifier. Under these circumstances the quiescent power consumption becomes a much bigger percentage of the overall consumed power when compared to the actual power delivered to the load.
The concept of "power average back-off" arises as a means to describe the practical operating point that permits linear distortion-free transmission. Typically, additional back-off is also required to ensure that the transistor is not driven into saturation, where it becomes very non-linear and creates substantial spectral regrowth. This incurs a further loss of efficiency.
Historically, this back-off has not been a big concern in OQPSK or QPSK satellite systems, where the modulation exhibits a modest 3dB crest factor. This slight back-off, in turn, still yields efficiencies of around 30%, which were often deemed acceptable. Unfortunately, the advent of WCDMA and multi-carrier WCDMA has lead to information-bearing wavefor ms that often exhibit crest factors in excess of 10dB. Backing off a Class A or AB amplifier to operate within its linear operating region with this kind of waveform rapidly forces highly inefficient operation. Under these operating conditions typical efficiencies of less than 10% are observed.
The topology of predistortion offers efficiency enhancements because, unlike a feed forward amplifier, a second energy-wasting error amplifier is not required. However, predistortion offers further incremental efficiency gains. These gains are extracted by remembering that the predistortion kernel develops a baseband non-linearity that is complimentary to the entire amplifier characteristic. This permits the back-off requirement to be minimized because additional margin need not be sacrificed to avoid unwanted distortion that is associated with operating near the saturation and 1 dB compression point of the amplifier.
Basically, predistortion provides the correction factor that permits using the ampli fier right up to the saturation point. By realizing that the maximum signal crests within a multi-carrier system occur very rarely, more aggressive efficiency gains can also be achieved. The amplifier back-off or operating point may be adjusted to a more efficient point, so that on these rare occurrences the amplifier is actually overdriven deep into saturation. This event can never be compensated for in a predistortion system because no amount of correction will enable the amplifier to deliver more power than it is capable of generating.
During the overdrive event the distortion that is generated results in very high instantaneous spectral regrowth, however because of its very infrequent nature the energy contribution to the average power spectral density remains negligible. Astute system operators will, in fact, deliberately overdrive predistortion systems to extract this increased efficiency, knowing that any signal crest that has a probability of occurrence that is less than 10-4 will not measur ably degrade the systems average power spectral density.
PMC-Sierra's Paladin predistortion system has been developed to permit these aggressive efficiency strategies while maintaining absolute system stability. Using these combined approaches, efficiencies up to 20% can be readily achieved with WCDMA multi-carrier systems.
Essentially, the crest factor of an information-bearing waveform has a profound effect upon the amplifier efficiency. Thus, there is clear motivation to explore techniques that dramatically reduce the crest factor of single and multi-carrier WCDMA waveforms.
See related chart