R. Neil Braithwaite

Wide bandwidth adaptive digital predistortion of power amplifiers using reduced order memory correction

Adaptive digital predistortion (DPD) with memory correction is made challenging when the observation bandwidth used to measure the output signal is narrower than the discernable intermodulation (IMD) products generated by the input signal and power amplifier nonlinearity. This is likely to occur when the input signal has a wide bandwidth and the PA has severe nonlinearities, as in the case of a Doherty PA. Wide bandwidth signals tend to stimulate the memory effects within the PA requiring memory correction to be included in the DPD. However, if in the pursuit of additional memory correction the number of basis waveforms used is over-specified relative to the measurable output signal bandwidth, the estimation of the memory coefficients will be ill-conditioned. This paper proposes a method based on eigenvalue decompositions to reduce the order of the memory coefficient estimation. It also shows how measurements from multiple narrow bandwidth observations can be combined to increase the available rank of the estimation and improve the overall cancellation performance. Experimental results obtained using a RF test bench comprising a Doherty PA and a 20 MHz input signal demonstrate the relationship between the available order of the estimation, the number of observation bandwidths used, and the performance of the wide bandwidth IMD cancellation.

Digital predistortion of a power amplifier for signals comprising widely spaced carriers

Digital predistortion (DPD) of a RF power amplifier (PA) for signals comprising two widely spaced carrier clusters is proposed. The basis waveforms within the DPD are selected to allow for the use of a lower sampling rate. Multi-carrier signals spanning 60 MHz of bandwidth are linearized using a sample rate of 100 MHz for the transmitter and observation paths, as oppose to a sample rate exceeding 300 MHz (5 times Nyquist) used conventionally. A transmit (Tx) filter located after the PA attenuates intermodulation distortion (IMD) falling outside of the 100 MHz of correction bandwidth provided by the DPD.

Reducing estimator biases due to equalization errors in adaptive digital predistortion systems for RF power amplifiers

This paper investigates degradations in a RF power amplifier, linearized using digital predistortion (DPD), due to biases in the DPD coefficient estimation. Of specific interest is the effect of equalization errors in the observation path on the distortion spectrum of the RF output signal. It is proposed that the frequencies allocated to the input signal be notched within the estimator to reduce the effects of equalizer errors. Results show that the use of an estimator notch improves the ACLR2 performance by 11 dB and 16 dB for DPD implemented using memory and gain polynomials, respectively. The notch also allows the order (rank) of the estimation to be reduced.

A comparison of indirect learning and closed loop estimators used in digital predistortion of power amplifiers

Indirect learning is often used as an estimator in digital predistortion of power amplifiers (PAs). The estimator has inherent flaws that become apparent when signal bandwidths increase. These include coefficient offsets, excessive ADC sampling requirements, and susceptibility to EVM and PA saturation. A comparison to the closed loop estimator shows that these flaws are specific to the indirect learning estimator.

Memory correction for a WCDMA amplifier using digital-controlled adaptive analog predistortion

A digital-controlled adaptive analog predistorter, described in [3], is extended to accommodate wider bandwidth WCDMA signals and to provide memory correction. The desirable characteristics of the original system are preserved, which include low cost, low insertion delay, adaptive coefficient estimation, and a data acquisition system using sub-Nyquist asynchronous sampling. The predistortion function is based on a zero-crossing structure derived from the input signal envelope and its derivative. Memoryless distortion is corrected using the input envelope only whereas the memory correction also includes the derivative of the input envelope. A Matlab simulation modeling a class AB biased amplifier verifies the feasibility of wide bandwidth operation. Tests using a WCDMA input signal with a 101-carrier configuration indicate that memory correction can improve the ACLR2 by 2.7 dB, which is comparable to the results presented in [5].

Measurement and correction of residual nonlinearities in a digitally predistorted power amplifier

A reduced cost method for the measurement and correction of residual nonlinearities in a digitally predistorted transmitter is proposed. The original system [1] uses a calibration signal applied off-line and a cancellation loop with a square law detector to measure residual nonlinearities. The approach is improved by replacing the calibration signal with the actual signal transmitted, resulting in an on-line measurement system. A further extension integrates several measurements over the input envelope range to adapt a fourth-order polynomial predistorter. Simulation results for a WCDMA input with a 101 carrier configuration show that the proposed approach for measuring residual nonlinearities and estimating predistortion coefficients is effective.

High efficiency feedforward power amplifier using a nonlinear error amplifier and offset alignment control

A high efficiency feedforward power amplifier (FFPA) is proposed that uses the Doherty configuration for the both the main amplifier (MA) and error amplifier (EA). Using a nonlinear EA is a deviation from the traditional FFPA. A loop control algorithm is proposed that offsets the alignment setting from the standard value to best utilize the power handling capability and linearity of the MA and EA. A system efficiency of 25% is achieved for a rated output power of 52.7 dBm, while exceeding the WCDMA specifications. Reducing the peak-to-average power ratio (PAPR) of the input signal to 8 dB and using offset control improves the ACLR performance to -62 dBc.

Computationally efficient coefficient estimator used in memory-based digital predistortion (DPD) models

Computationally efficient estimation of digital predistortion (DPD) coefficients for memory polynomial, gain polynomial, and pruned Volterra series models is proposed. In all three cases, the basis waveforms within the estimator are specified in the frequency domain as a function of memoryless waveforms and delay operators, thereby reducing the number of FFTs needed and allowing the tap spacing to be a fraction of a sample. The gain polynomial and pruned Volterra series cases are approximations that are sufficiently accurate for a closed loop estimator to converge to a desired steady-state.

Implementing crest factor reduction (CFR) by offsetting digital predistortion (DPD) coefficients

A combined approach to digital predistortion (DPD) and crest factor reduction (CFR) is proposed. The new CFR is structured similar to DPD and is implemented by introducing a steady-state offset into the DPD coefficients. The DPD and CFR coefficients are estimated using separate adaptive processes but applied to the transmission path in a common module. Results show that CFR lowers the PAPR of the predistorted signal by 5 dB, which reduces the stress on the peaking transistor in a Doherty PA.

Echo Cancellation for a Wide Bandwidth Mixed-Mode WCDMA/GSM Repeater with Digital Sub-Band Filtering

This paper describes an echo cancellation (EC) algorithm used in a mixed mode WCDMA/GSM repeater with digital sub-band filtering. The EC coefficients are estimated iteratively from cross and power spectrums improving the computational efficiency of the estimation. The effect of regularization and spectral normalization on the convergence of an adaptive EC algorithm is investigated. Singular value decomposition (SVD) is applied to the estimation process as a form of regularization, compensating for the bandwidth reduction associated with the sub-band filter. Spectral normalization used in the estimation of the EC coefficients is shown to avoid convergence problems and the frequency- dependent cancellation that affect the baseline approach. Simulation results show echo cancellation of 23 dB across a 60 MHz bandwidth.