GB2357915A

GB2357915A - Correcting distortion in an all-digital power amplifier using transfer function data in a look-up table

Info

Publication number: GB2357915A
Application number: GB9930860A
Authority: GB
Inventors: John Francis Hope
Original assignee: Individual
Current assignee: Individual
Priority date: 1999-12-30
Filing date: 1999-12-30
Publication date: 2001-07-04
Anticipated expiration: 2019-12-30
Also published as: GB2357915B; GB9930860D0

Abstract

A scheme for correcting distortion arising in the power switching stage of an all-digital power amplifier is disclosed. The invention allows low distortion to be obtained from amplifiers using simple, cost-effective power switching stages. A stored transfer function (5) characterizing the power switching stage (3) and associated reconstruction filter (4), provides the information means for the amplifier's digital modulation means (2) to pre-compensate for non-linearities in the power stage and reconstruction filter. The data values constituting the transfer function are specific to each individual amplifier and may be obtained by an automated series of calibration measurements at production time.

Description

2357915 A SCHEME FOR CORRECTING DISTORTION IN AN ALL-DIGITAL POWER

AMPLIFIER

BACKGROUND.

Audio systems are becoming progressively more digital and one of the remaining analog links in the sound reproduction chain is the power amplifier. Audio is readily stored and processed digitally, but loudspeakers are intrinsically analog devices and require an analog voltage to drive them. Conventionally the digital audio samples are converted to a low voltage analog audio signal by a DAC (Digital to Analog Converter) and then amplified in analog form in order to drive a loudspeaker. However, high power low distortion analog power amplifiers are inefficient, bulky and prone to the vaguaries of component tolerence and drift.

Class D power amplifiers with switching output stages offering typical efficiencies of over 90% were first developed in the 1960's but were plagued by unreliability and high switching losses. Modem switching devices such as MosFets, and the advances made in integrated circuit engineering have recently led to renewed development in Class D power amplifiers. Hitherto these amplifiers have accepted analog input signals, which implies unnecessary digital to analog conversions in the reproduction chain. The goal is to eliminate these redundant conversions and produce an all digital power amplifier.

An all-digital audio power amplifier is a power amplifier which accepts digital audio samples as input, processes these entirely in the digital domain to control a switching output stage which can deliver audio power to a loudspeaker. Such amplifiers have been developed as university research projects by Sandler, Pedersen and others and the first commercial unit was launched in 1998.

OPERATION OF AN AL1,1)1GITAL POWER AMPLIFIER Figure 1 shows a simplified block diagram of a typical all-digital power amplifier and represents prior art.

An all-digital power amplifier accepts as input digital audio samples in a format such as established by the Audio Engineering Society or other industry standard. The incoming audio samples have low sampling rate and high resolution - 44. 1 kHz sampling rate 24-bit samples being typical and are input to an oversampling means (1). The oversampling means converts the lowsampling rate, high resolution input samples into medium sampling rate, high resolution samples with redundant information capacity, typically 8 times oversampled 24-bit samples at 352.8kHz. These are input to a digital modulation means (2) which utilizes the redundant information capacity to convert the medium sampling rate, high resolution samples to a medium sampling rate, low resolution pulse stream suitable for driving a power switching stage.

A fundamental requirement of the modulation is that the frequency spectrum of the medium sampling rate low resolution pulse strewn should contain the true analog equivalent signal at baseband, ie from dc to 20kHz. Viable modulation types include Pulse Width Modulation (PWM), in which'width of pulses occuring at fixed repetition rate is varied, or Delta-Sigma Modulation in which the pulses have fixed width and the density of these pulses is varied. In Figure 1 the digital modulation means (2) is of a typical Pulse Width Modulation (PWM) type and comprises subsections as indicated. (21) is an Intersector employed to linearize the Uniform PWM (UPWM) modulator (24). (22) is a quantizer which requantizes the 8x oversampled 24-bit audio samples into 8-bit samples at 352.8kHz, speetrally shaping the resulting quantization noise by means of a digital filter (23) so that it is effectively 'dumped' outside the audio band. The 8-bit _nise-shaped audio samples output by (22) are converted into 256 level PWM by the UPWM -.,.:bdulator (24).

The output from digital modulation means (2) controls a power switching stage (3) wlich regenerates the pulse stream at sufficiently high voltage and current capacity to produce the desired output power. A typical power switching stage may comprise a full bridge of MosFet power transistors configured to generate a bipolar, binary (often called Class AD modulated) power pulsetrain. A passive lowpass reconstruction filter (4) passes the baseband audio in the power pulsetrain to the loudspeaker but filters out the undesired out-of-band high frequency moc. ulation components.

bISTORTION ARISING IN THE POWER STAGES OF ALL-DIGITAL POWER AMPLIFIERS It will be observed from Figure 1 that there is no overall feedback path. The input to the a plifier is digital and the output to the loudspeaker is analog, thus any overall feedback scheme would necessitate the inclusion of a means of analog to digital (A/D) conversion, thereby negating the all-digital aspect of the design. Thus the amplifier is inherently open-loop, which places critical demands on the switching performance of power switching stage (3) if low distortion is to be obtained. An ideal power switching stage would generate perfectly rectangular pulses of exactly constant amplitude, zero transition time and zero source impedance. However, in any p -actical implementation the pulse shape is not perfectly rectangular and this results in distortion. Pulse distortion in a typical power switching stage is caused by, inter alia:

Finite and unequal switching rise and fall times of the power switching devices.

'Dead time' when both switching devices in a phase arm are off. The introduction of these delays is necessary to avoid shoot-through which can destroy the devices, and tz) keep reverse recovery currents within safe limits.

Parameter spread in rise/fall time, delay time, and on state resistance of the power sm itching devices.

The forward voltage drop across the freewheel diodes which follows a square law.

Parasitic circuit elements causing ringing, overshoot and other artifacts on the pulse s iape.

Distortion arising in the power stages usually dominates the overall distortion perforinanc( of an all-digital power amplifier and is difficult to correct without introducing feedback.

A SCHEME FOR CORRECTING DISTORTION ARISING IN THE POWER STAGES OF ALLDIGITAL POWER AMPLIFIERS If it is assumed that the nonlinearities introduced by the power switching stage are time invariant, repeatable, and the output is a function of only the applied PWM value and not of p vious applied values, it follows that it would be possible to characterize the transfer function of i given power switching stage and devise a scheme which pre- compensates for distortion introduced by that power stage. Persons invested with skills in the art will appreciate that the parameter spread across power switching devices alone would preclude mathematical modeling of such a iransfer function to any degree of useful accuracy.

However, the invention to be described circumvents the mathematical intractibility and utilizes a set of measured in-out values unique to each individual power switching stage that coffi pletely characterize the transfer function. These values are measured during a calibration routine at production time and the scheme does not add any AJ1) conversion, feedback, or increased component cost to the amplifiers themselves.

2.

fer to Figure 1. The input to the filter (23) is solely the error due to the requantization by quantizer (22). The implicit assumption is that the UPWM modulator (24), power switching stage (3) and passive lowpass reconstruction filter (4) will together accurately reproduce the analog equivalent of each 8-bit value output by the quantizer (22). As has been discussed above this is not the case and distortion results because in this configuration the noise-shaping quantization system is unable to correct for non-linearities occuring downstream of the quantizer.

Refer now to Figure 2, which shows a block diagram of an all-digital amplifier incorporating the invention. A means of stored transfer function (5) of the power switching stage and passive lowpass reconstruction filter has been added. (5) contains as many entries as there are output states of quantizer (22). Each entry consists of an (IN, OUT) pair such thatfor each quantizer state there is a corresponding digital equivalent of the analog voltage produced at the output of the amplifier when that particular quantizer output is applied to the UP1W modulator (24). These digital equivalent values must be represented with high precision. In practice the resolution is conveniently set to the base wordlength of the computational means employed typically 24 bits - and the means of stored transfer function (5) is implemented as a Lookup Table (LUT) in non-volatile memory. It can be seen that the noise shaping quantization system is now able to compensate for errors occuring downstream of the quantizer (22) because the means of stored transfer function (5) provides a value of the actual output of the amplifier, from which a true quantization error signal may be derived and fed back into the digital filter (23).

The contents of the means of stored transfer function may be obtained from a sequence of calibration measurements taken during factory testing, or whenever otherwise deemed necessary. Calibration requires an external A/D converter of high precision, (eg 24-bit instrumentation quality ADC with high dc accuracy spec), connected to the amplifier output, and a means of controlling the calibration sequence which may be internal or external to the amplifier. The UPWM modulator (23) is stimulated through all 256 possible inputs and for each stimulus, the digital output from the external A/D converter is read and stored in the non-volatile lookup table which constitutes a means of stored transfer function (5).

The scheme for correcting distortion described above has been implemented in an all digital power amplifier. Measurements showed that an effective reduction in Total Harmonic Distortion of well over an order of magnitude was achieved. The benefit of the invention can be realized in two main ways:

All-digital power amplifiers which employ costly and esoteric semiconductors in 'nocompron-tise' power switching stages can realise improved distortion performance.

All-digital power amplifiers can be designed around simple, costeffective power switching stages and still deliver good performance.

S.

Claims

A power amplifier having digital input and analog output and incorporating a digital modulation means and a power switching means and a lowpass filter means, and which further incorpc rates a means of stored transfer function of the power switching means and lowpass filter means.

(4_.

Amendments to the claims have been filed as follows Claim:

A power amplifier receiving a digital input signal and providing an analog output signal and incorporating:

a) a means of oversampling the digital input data, and b) a means of requantising the oversampled data output from means (a) and which additionally determines the resulting quantisation error and shapes the frequency spectrum of the quantisation error, and c) a means of digital modulation operating on the oversampled, requantised data output from ineans (b) and d) one or more power switching devices and a means of controlling these devices by the digitally modulated output of means (c) and e) a means of low pass filtering operating on the switched signal produced by means (d) and from which the analog output of the amplifier may be derived, and f) A means of storing values representing the transfer function of the switching devices (d) or the combined transfer function of the switching devices (d) and the lowpass filter means (e), such that:

The means of transfer function storage (f) is used together with the requantising means (b) in such a manner that values obtained from the means of transfer function storage (f) are used in determination of the quantisation error by means (b).

1/1 s-