CN1151606C - Wave filter for digital to analog converter - Google Patents

Abstract

An analog-to-digital converter responsive to an analog signal converts the analog signal to a digital signal. A digital decimation filter, having an associated frequency response, is responsive to the digital signal for producing a decimated digital signal. An alias correction filter attenuates the decimated digital signal to remove distortion. A digital interpolation filter, having an associated frequency response, is responsive to the alias-corrected signal for producing an interpolated digital signal which subsequently is converted to an analog signal by a digital-to-analog converter. The alias correction filter has a frequency response that attenuates the decimated digital signal in a frequency range generally corresponding to a distortion portion that is present within the combination of the digital decimation filter frequency response and the digital interpolation filter frequency response.

Description

The method that is used for the filter and the reduction aliased distortion of digital to analog converter

Technical field

Improvement in relate generally to digital-to-analogue of the present invention (D-to-A) conversion, and be particularly related to improvement for this conversion of audio signal.This high quality audio that is used for music playback in such as mini disk (CD) player, DVD player or the like device is very useful.

Background technology

The modulus (A-to-D) in the modern times of the audio frequency that is useful on and digital to analog converter are worked being higher than on the sample rate of output/input sampling rate basically.They are called over-sampled converter, and their use digital filter to select under the situation of analog to digital converter to be reduced to output sampling rate, and carry out interpolation from the sample rate of input under the situation of digital to analog converter.An important reasons of this structure is, in order to reduce the distortion that causes by undesirable frequency that exists in modulus and the digital-to-analogue conversion process as far as possible, filter is necessary, and digital filter is more more stable than the analog filter of equal quality, more can reset, and it is lower to implement cost.Fig. 1 a is the block diagram that analog to digital converter is simplified, and Fig. 1 b is corresponding digital to analog converter.

The actual sample rate of transducer itself may be than much higher times of I/O sample rate in the over-sampling digital quantizer.Select or what interpolation process normally carry out in, last decimation filter and first interpolation filter constitute with two-to-one frequency ratio usually.Cao Zuo these filters have best effect on the sound performance of audio converter at last/first, because their cut-off frequency is near the frequency in the program material.Fig. 2 a is illustrated in the frequency response of the typical filter that is used as last decimation filter in the analog to digital converter, and Fig. 2 b is illustrated in the response of the typical filter that is used as first interpolation filter in the digital to analog converter.The Y-axis of figure is represented the value that the filter amplitude responds with decibel, and X-axis is the mark (f of frequency representation for output/input sample frequency Fs _K).The reason that detects modulus and two kinds of filters of digital-to-analogue is that their function is the system as the effect of determining several distortions in the output signal 180.

The mark 0.5 of the centre of figure is corresponding to significant so-called nyquist frequency.Its important reasons is that sampling thheorem claims that sample frequency frequency over half can not be represented by the data flow that is sampled uniquely in the data system that is sampled.Under the situation of analog to digital converter, any frequency more than the nyquist frequency of not removed by decimation filter in the input signal will occur as the parasitic frequency in the output, be called aliasing frequency or aliased distortion.From the viewpoint of aliased distortion, desirable decimation filter will make all frequencies that are lower than 0.5Fs pass through, not the above energy of 0.5Fs frequency.In fact this filter is irrealizable, and actual filter is attempted the desirable response of convergence usually.The above any residual frequency of Nyquist is overlapped in or is aliased in the following frequency of Nyquist in the output signal 120 in the primary signal, and its relation is that the frequency f in the input becomes the Fs-f in the output.

Aliased distortion mechanism also is present in the interior interpolation process of digital to analog converter.Supplied with digital signal 122 among Fig. 1 b can be considered to not have the above frequency of 0.5Fs.The first order of interpolation is by adding the null value sample so that sample rate doubles between each original sample, and then the result transmitted by the low pass filter that has as the response among Fig. 2 b.Consequently, the null value sample is replaced by the value from the ambient data interpolation.

Distortion is to be caused by the following fact, and new frequency is to generate more than original nyquist frequency, and these new frequencies are corresponding to the frequency that occurs in the primary signal.In order to analyze the potential influence of this distortion, with graphical presentation select/the combination frequency response of interpolation system is useful.If get the frequency response of the modulus decimation filter among Fig. 2 a, and selecting of being equal to, before interpolation filter, insert the null value sample subsequently, obtain the response shown in Fig. 3 a.To each frequency under the Nyquist 200, generate the above new frequency of Nyquist.From about the symmetry of Nyquist 200 as seen, these have the relation that identical f_ that aliases products had newly equals FS-f under the situation of modulus.These new frequencies are by unique expression, because sample rate is that twice is big now.

If add the cascade of the interpolation filter response of Fig. 2 b now, then obtain the array response shown in Fig. 3 b.0.5 above frequency is the signal that does not have in primary signal, and is the product of aliased distortion.They are divided into two groups greatly: be marked with 220 signal corresponding to the stopband of interpolation filter, and be marked with 210 with select the signal relevant with the transition band behavior of interpolation filter.

Many people think that these distortion product are unessential, because their above unnecessary signals that is useful bands, and are unheard at CD or any sample rate in other system greater than 40kHz.If each thing all really is linear in following the audio system of interpolation filter, this will be correct.Regrettably, the Shi Ji world is not strict linearity.Digital to analog converter, small signal amplifier, power amplifier, loud speaker and even people's the sense of hearing in all exist non-linear.

Among Fig. 3 b, acceptable distortion product level 220 has determined the Stopband Performance requirement to interpolation filter.The unique channel that reduces these distortions is to improve the performance of interpolation filter stopband rejection.

The distortion at 210 places is subjected to bigger restriction on frequency range, but they have higher amplitude, and can cause the actual problem of hearing in the output of system.As an example, consider to produce in the music sound of the cymbals of large amplitude radio-frequency component.For each composition under the nyquist frequency, of the above border of nyquist frequency picture frequency rate correspondence is arranged, and each of original frequency and aliasing frequency will be to producing the difference frequency composition when running into non-linear in the system afterwards.In the situation of the CD system that has these filters, these difference frequency compositions are in the frequency range of 0 to 5 KHz, and this sense of hearing to the people is very sensitive, and wherein they are not sheltered well by the signal that produces them.Consequently produce the sound of cymbals " dirt ", this is a Typical Digital system very.

These 210 type distortions result is mainly derived from the transition band behavior of interpolation filter.Be generally used for that the filter type of this position is called half-band filter in the system design.As from Fig. 2 b finding, the decline of 6dB is arranged at the nyquist frequency place, suitable response is arranged more than the nyquist frequency.This uses in many systems, because it implements very economical on calculating, and because good time domain performance is arranged.This is finite impulse response (FIR) (FIR) filter that has the symmetry of linear phase response, and it strictly is zero that wherein all even-order coefficients are removed outside middle one, thereby needn't make those multiplication.The filter of this type is used on the digital to analog converter of design for the most commercial of audio frequency use.

At United States Patent (USP) #5,479,168 and #5,808,574 and associated materials in revealed the prior art method that solves with half frequency band interpolation filter relevant issues.Be to use the interpolation filter that begins to weaken in stability at lower frequencies from the solution of the viewpoint optimization of performance, rather than half band filter.By the filter of use with employed decimation filter complementation in analog to digital converter, can sharply reduce the amplitude of transition band aliased distortion composition, still keep good time-domain pulse response simultaneously.Can also control to Stopband Performance on any required level.In a lot of commercial application, this method has two main shortcomings.

First shortcoming is a cost.The complexity of non-half-band filter that is applicable to this application is usually above the twice of tradition half band filter of correspondence.This directly causes the larger-size yardstick of physically realizing the needed semiconductor chip of filter, and thereby hard-wired expensive.

Second shortcoming be, the output sampling rate of interpolation filter is the twice of the Fs of input preferably, and the digital to analog converter of following filter must can be accepted higher sample rate.Because the many transducers that use in modern comfort are actually junction filter/digital to analog converter, so they can not accept higher sample rate.

Summary of the invention

The present invention has eliminated a kind of aliased distortion that the transition band behavior of half frequency band interpolation filter commonly used causes in the audio digital to analog converter.The present invention realizes this point and does not change interpolation filter itself.When suitably designing, it can be eliminated aliased distortion and not have the obvious reduction of system's time-domain response.

According to an aspect of the present invention, the digital filter that together uses with the digital to analog converter with sample frequency Fs is provided, this transducer is the type with half relevant frequency band interpolation filter, this digital filter places the inlet flow of transducer, and comprise a low pass filter, described low pass filter weakens the frequency in the frequency band that comprises nyquist frequency in the input signal, makes the filtering of being undertaken by described half frequency band interpolation filter subsequently present the transition band distortion that reduces.

According to a further aspect in the invention, a kind of method that reduces aliased distortion in the audio digital to analog converter is provided, comprise: the inlet flow of logarithmic mode transducer carries out filtering to weaken the frequency that falls into the frequency band that comprises nyquist frequency in the input signal, makes the filtering of being undertaken by described half frequency band interpolation filter subsequently present the transition band distortion that reduces.

Description of drawings

Fig. 1 a is the block diagram that analog to digital converter is simplified.

Fig. 1 b is the block diagram of the digital to analog converter corresponding with the analog to digital converter of Fig. 1 a.

Fig. 2 a is illustrated in the frequency response that is used as the typical filter of last decimation filter in the analog to digital converter of Fig. 1 a.

Fig. 2 b is illustrated in the response that is used as the typical filter of first interpolation filter in the digital to analog converter of Fig. 1 b.

Fig. 3 a is the frequency response of getting the modulus decimation filter among Fig. 2 a, and selecting of being equal to, and inserts the response that the null value sample obtains subsequently before interpolation filter.

Fig. 3 b is the array response that the cascade of the interpolation filter response of interpolation Fig. 2 b obtains.

Fig. 4 illustrates the diagrammatic sketch that comprises according to the digital-to-analog conversion system of filter of the present invention.

Fig. 5 a illustrates the frequency response according to the exemplary of filter of the present invention.

Fig. 5 b illustrate when according to the response of filter of the present invention with previous Fig. 3 b select with interpolation filter cascade classification the time, the response that obtains.

Fig. 6 a illustrates the curve that makes up with interpolation filter of selecting of Fig. 3 b.

Fig. 6 b illustrates the example that has added the response that the aliasing correcting filter had.

Fig. 7 illustrates a kind of possible distortion according to filter of the present invention.

Fig. 8 illustrates another example according to filter of the present invention.

Fig. 9 illustrates according to prior art, obtains the example of HDCD system form of the present invention's advantage through modification.

Figure 10 is another example that is similar to the HDCD system of Fig. 9.

Embodiment

The solution of implementing among the present invention to this problem is before half frequency band interpolation filter, has added another filter to D/A system.For following discussion, referring to Fig. 4.

Original digital-to-analog converter structures is shown in among dotted line 140 encirclements, and in a lot of commercial application, such several passages are integrated in the single integrated circuit (IC).The filter 130 that the present invention adds places the inlet flow to original transducer.Be noted that importantly 130 input 122 is in identical sample rate Fs with output 135, be called the reason of 1xFs aliasing correcting filter and Here it is.

The frequency response of the exemplary of this filter is shown in Fig. 5 a.This is the low pass filter that only weakens in the input signal very near the frequency of Nyquist.(for example from about 0.45 to about 0.55Fs).The curve representation of filter is prepared to carry out interpolation, so that its relation with other curve is clear for being inserted with zero sample.The example of shown filter is the FIR filter with symmetry of linear phase response.Because it is operated in Fs, so in order to stipulate that its behavior only need be up to the frequency response of 0.5Fs.

When the response of this filter with previous Fig. 3 b select with interpolation filter cascade classification the time, obtain the response of Fig. 5 b.Notice that the transition band aliasing response 210 among Fig. 3 b is subjected to effective inhibition in Fig. 5 b, eliminated the aliased distortion that the use by half band filter causes.

Make corresponding to the aliasing response of the stopband of interpolation filter constantly by adding filter,, can not change the zone of that response because be operated in the performance of filter in keeping frequency band of Fs.Unique way that can solve the stopband relevant issues is to use with the filter than high sampling rate.

Listen to test and point out, the transition band aliased distortion is recently from the higher-frequency of stopband range, more can hear than low amplitude distortion.The introducing of filter 130 (Fig. 4) can significantly improve the sound of system.

In order to suppress the aliased distortion product in the interpolation filter transitional region, should be suppressed at some frequency at the top of the normal response of primal system.The filter 130 that adds is low pass filters.Having CD or more in the system of high sampling rate, inessential on acoustics usually in the loss of about 20 KHz and higher frequency.Also have another to sound as if the phenomenon of high frequency loss of sharpness, this phenomenon is actually by different mechanism and causes, and is promptly relevant with transient energy dispersion in time.We are referred to as time ambiguity.

If get pair upright leaf inverse transformation of the grade frequency response among Fig. 3 b, then obtain the impulse response of hierarchy system equivalence.From hearer's viewpoint, this is more even more important than filter impulse response separately, is the combination of filter because in fact signal pass through.If investigate the classification impulse response of regular linear-scale, look that then the example that resembles very much indivedual filters responds.Regular linearity curve does not have to disclose the information of a lot of relevant sense of hearing experiences.

If yet draw the impulse response value with logarithmic scale, can see some absorbing result, this with listen to test relevant.Human auditory's response is to numerical expression.Fig. 3 b is this simply to select curve plotting with interpolation filter combination in Fig. 6 a for resembling.Vertical axis is the amplitude of representing by dB with reference to full size, and horizontal axis is by sampled representation.This can be envisioned as system by the response of time to the single pulse of unit amplitude (0dB).Notice that system is dispersed on many samples in time to the energy in the signal pulse peak response.If supposing this is that sample frequency is the example of the CD of 44.1kHz, be that the point of the following 50dB of full size is in the distribution time and is approximately 3 milliseconds then at curve.This time is corresponding to the about one meter physical size of the velocity of sound that provides in the air.If source of sound is little originally, such as the wooden unit percussion instrument, then time distribution or time ambiguity are changed into sound by system and are reset in the mode that sound has seemed to lose the high frequency definition.

When adding 1xFs aliasing correcting filter, just added another impulse response to cascade.Owing to this equates the convolution of the impulse response of the impulse response of new filter and previous combination, so the length overall of impulse response is bigger.This needn't mean that also sound is relatively poor.The part of most important response curve is the above zone of approximately negative 80dB on the acoustics.As if can take a kind of Filter Design, it has less time distribution in higher amplitudes, and at very low amplitude more distribution is arranged, and this sounds having high frequency definition preferably than the system that does not add the aliasing correcting filter.Example with this response is shown in Fig. 6 b.

Among curve among Fig. 6 b and Fig. 6 a is that identical selecting/interpolation filter is right, with the aliasing correcting filter classification of Fig. 5 a.Note, among Fig. 6 b curve-width more than the 75dB is narrower than the part of Fig. 6 a correspondence.In fact the combination of the filter among Fig. 6 b sounds than Fig. 6 a having high frequency response preferably, even in fact it have less radio-frequency component because of low pass filter.Because suppressed the aliased distortion from transitional region, it sounds the distortion much less.

When these aliasing correcting filters of design, it is important to check the result in frequency domain and the time domain, because two kinds of viewpoints are represented the aspects that acoustic efficiency designs.Two kinds of viewpoints are different usually each other.Such as this problem of aliased distortion, be easy to solve by introducing another time distribution.Last selection usually is must be by listening to a kind of balance that test solves.

The combination of 1xFs aliasing correcting filter and half frequency band interpolation filter and digital to analog converter usually implements the non-half band filter method more complicated than prior art and comes economically.In many contemporary audio assemblies that use for the consumer, there is a kind of nextport universal digital signal processor NextPort (DSP) function to add the integrated filter/digital to analog converter that uses half band filter.Example has and uses the DSP function to the compressed audio decoding and be used for DVD player and A/V receiver such as other functions such as bass managements.Usually be that enough extra process abilities are arranged in the DSP function that realizes in this system,, except that the program ROM that is used for DSP, do not change any hardware so that add the aliasing correcting filter to system.The present invention allows system designer to use the integrated filter/digital to analog converter of identical standard, and obtains that the advantage of hanging down the better quality conversion of aliased distortion is arranged.Among Fig. 7, the DSP function is by representing with the dotted line of label 131 marks.The DSP function can be by also realizing that with the integrated device of digital to analog converter 140 perhaps the DSP function can be realized by the device that separates.Independent DSP device is known in prior art, and the DSP function can be realized with analog D SP function such as the microprocessor with abundant computing function by other device realization really.Being used for controlling the DSP functional programs is stored in the program ROM 132 usually.

Use filter to be considered to new with the input sample frequency of digital to analog converter with novel to reduce its aliased distortion.As if the character of transition band aliased distortion do not understood in current document well, and except using HDCD, not reflection in the current system design.And classification set forth above select/frequency domain of interpolation filter system and time domain behavior analysis method and the application in design interpolation filter or aliasing correcting filter thereof are seemingly unique.

So far example letter has covered simple situation with interpolation filter and the direct-connected aliasing correcting filter of digital to analog converter.Be apparent that basic structure of the present invention has many possible distortion with identical basic function.A kind of possible distortion is shown among Fig. 7.Under this situation, aliasing correcting filter 130 is inserted in the data flow at input 122 places, before signal enters interpolation filter 150, can carry out other digital processing 137 to it then.Need not to change essence of the present invention, this other processing 137 can comprise bass management, Space, tone control, reverberation or the like, and can be realized well under the program control of ROM 132 by the DSP function as mentioned above.Because behavior that aliasing is proofreaied and correct is based on the graded filter of system, all is feasible so add other processing Anywhere in chain, can not change the basic act of correct effect.

Another example is shown among Fig. 8.This is wherein to use the example of some form of data compression with the system of realization higher data storage density.The example of this system is AC3, mpeg audio compression etc.The part that is specifically designed to the playback system 125 that compressed format is decoded or decompressed must be prior to aliasing correcting filter 130.Before aliasing correcting filter 130, this module also can comprise other processing.The output of aliasing correcting filter 130, directly or as need be applied to the digital interpolation filter by other Digital Signal Processing 137.As the embodiment situation of formerly discussing, aliasing correcting filter 130 is preferably realized by DSP, and is more preferably by being used for realizing that the same DSP of other Digital Signal Processing (this processing if sample) realizes.

Fig. 9 illustrates according to the prior art of contrast described in the patent, obtains the example of HDCD system form of the present invention's advantage through modification.In this system,, be used for the decimation filter (not shown) of encoder for best fidelity Dynamic Selection according to the content of program material.Both optimize owing to the design of aliasing correcting filter (130) is based on employed decimation filter and interpolation filter 150, and because the selection of decimation filter sends regenerator to by the code edge channel of hiding, so can Dynamic Selection aliasing correcting filter 130, so that provide the best realization of decimation filter in any given time.This finishes by module 128, and the code information that this module recovery is hidden to being included in the command decode in the covered code, and uses control signal 129 to select one of several aliasing correcting filters 130.

Figure 10 is another example that is similar to the HDCD system of Fig. 9, has increased the decoding of HDCD amplitude in module 125.This amplitude decoding function is also controlled by code commands decoder 128 and control signal 126 by the covered code edge channel.In this system, the decoding of amplitude must be carried out before the aliasing correcting filter, so that make decoding correctly follow the tracks of coding.

Though preferably based on the symmetrical FIR filter of linear phase characteristic, the filter of other type also can be used to realize identical target to above-mentioned aliasing correcting filter.A kind of interesting filter is minimum phase (infinite impulse response) filter.This filter has such time domain specification, promptly occurs the filter ring after the transient affair that it causes.This has better time domain behavior from the audibility viewpoint for this, though near the phase shift the band edge may cause other realizable problem.

The example of the filter that uses in so far discussing be designed for CD Player and other sample rate 40 and 50kHz between system.The present invention also can be used for having the high resolution system than high sampling rate, such as allowing sample rate to be up to the DVD audio frequency of 192kHz.Because sample rate increases by eight or two, transition passage aliasing mistake problem remains a problem.The hemichannel interpolation filter remains the standard to the design of most of audio systems, and signal frequency composition-aliasing frequency composition still can cause the difference frequency in the voiced band during to non-linear sinking in running into system.Even the amplitude at the transitional region sound intermediate frequency signal of the filter of higher sample frequency is lower, this also is right.We 88.2 and 96kHz carried out listening to test, wherein transition band aliasing composition causes that the digital signal sounding is very loud, when with also be signal relatively the time, this sound is well as if the higher level of a lot of high frequencies arranged.Introduce the aliasing correcting filter to system and cause that the loud sound of being exaggerated disappears, the result is the very natural playback of original sound.

From above obviously as seen, though specific forms of the present invention is explained and described, can make various modifications not deviating under the spirit and scope of the present invention.

Claims (22)

1. the digital filter that together uses with digital to analog converter with sample frequency Fs, this transducer is the type with half relevant frequency band interpolation filter, this digital filter places the inlet flow of transducer, and comprise a low pass filter, described low pass filter weakens the frequency in the frequency band that comprises nyquist frequency in the input signal, makes the filtering of being undertaken by described half frequency band interpolation filter subsequently present the transition band distortion that reduces.

2. the digital filter of claim 1, wherein low pass filter only weakens the frequency that falls in this frequency band, and wherein said frequency band is about the nyquist frequency symmetry.

3. the digital filter of claim 1, wherein low pass filter only weakens 0.45 frequency in the 0.55Fs scope.

4. the digital filter of claim 2, wherein low pass filter only weakens 0.45 frequency in the 0.55Fs scope.

5. one of any digital filter of claim 1 to 4, wherein low pass filter is the symmetric finite impulse response filter.

6. one of any digital filter of claim 1 to 4, wherein low pass filter is realized by programmed digital signal processor.

7. the digital filter of claim 6, wherein digital signal processor is also realized other Digital Signal Processing effect, selectively comprises LFC low-frequency correction, signal decompression, reverberation, Space.

8. one of any digital filter of claim 1-4, wherein digital filter comprises a plurality of low pass filters with different qualities, and these filter response one command signals are so that select one of described a plurality of low pass filters that the inlet flow of transducer is carried out filtering.

9. the digital filter of claim 5, wherein digital filter comprises a plurality of low pass filters with different qualities, and these filter response one command signals, so that select one of described a plurality of low pass filters that the inlet flow of transducer is carried out filtering.

10. the digital filter of claim 6, wherein digital filter comprises a plurality of low pass filters with different qualities, and these filter response one command signals, so that select one of described a plurality of low pass filters that the inlet flow of transducer is carried out filtering.

11. digital filter as claimed in claim 7, wherein digital filter comprises a plurality of low pass filters with different qualities, and these filter response one command signals, so that select one of described a plurality of low pass filters that the inlet flow of transducer is carried out filtering.

12. the method for aliased distortion in the reduction audio digital to analog converter comprises:

The inlet flow of logarithmic mode transducer carries out filtering to weaken the frequency that falls into the frequency band that comprises nyquist frequency in the input signal, makes the filtering of being undertaken by described half frequency band interpolation filter subsequently present the transition band distortion that reduces.

13. the method for claim 12, wherein scope be digital to analog converter sample frequency 0.45 to 0.55Fs.

14. the method for claim 12, wherein filter step is realized by programmed digital signal processor.

15. the method for claim 14, wherein digital signal processor decompresses to inlet flow.

16. the method for claim 14 or 15, wherein digital signal processor increases Space to the audio frequency of being represented by inlet flow.

17. the method for claim 14 or 15, wherein digital signal processor provides bass correction to the audio frequency of being represented by inlet flow.

18. the method for claim 16, wherein digital signal processor provides bass correction to the audio frequency of being represented by inlet flow.

19. any one method of claim 12 to 15, wherein filter step is realized by one of a plurality of digital filters, and and then comprise the step that responds a command signal so that select the inlet flow of one of described a plurality of digital filters logarithmic mode transducer to carry out filtering.

20. the method for claim 16, wherein filter step is realized by one of a plurality of digital filters, and and then comprise the step that responds a command signal so that select the inlet flow of one of described a plurality of digital filters logarithmic mode transducer to carry out filtering.

21. the method for claim 17, wherein filter step is realized by one of a plurality of digital filters, and and then comprise the step that responds a command signal so that select the inlet flow of one of described a plurality of digital filters logarithmic mode transducer to carry out filtering.

22. the method for claim 18, wherein filter step is realized by one of a plurality of digital filters, and and then comprise the step that responds a command signal so that select the inlet flow of one of described a plurality of digital filters logarithmic mode transducer to carry out filtering.

Images