GB2561902A

GB2561902A - Digital signal processing

Info

Publication number: GB2561902A
Application number: GB1706797.6A
Authority: GB
Inventors: Melinder Bjorn
Original assignee: Soundtrap AB
Current assignee: Soundtrap AB
Priority date: 2017-04-28
Filing date: 2017-04-28
Publication date: 2018-10-31
Also published as: GB201706797D0

Abstract

An automated adaptive anti-clipping method, for use with a series of digital signal processing (DSP) units 1, tests each DSP unit for an output that reaches or is greater than a predetermined threshold, such as a clipping threshold, for the respective unit. If the output is less than the threshold, the successive testing of other units in the series is continued. If the output reaches its threshold, a gain reduction procedure is initiated on a DSP unit in a negative feedback manner. The multistage gain reduction procedure may be carried out on the unit under test or a unit upstream of the DSP under test. The gain reduction may performed internal to the DSP or by using an attenuation unit 3 at an input of the DSP. The gain reduction may be a function of an amount by which the output exceeds the threshold. Modules 2 may take one or more data samples of the output and select or derive a value for testing, the value being a maximum, aggregate, or average value. Multiple DSPs may be sampled simultaneously but tested in sequence, and a specific DSP may be selected for gain reduction.

Description

(54) Title of the Invention: Digital signal processing

Abstract Title: Initiating gain reduction to avoid clipping in a series of digital signal processors in a chain when the output from a DSP exceeds a threshold (57) An automated adaptive anti-clipping method, for use with a series of digital signal processing (DSP) units 1, tests each DSP unit for an output that reaches or is greater than a predetermined threshold, such as a clipping threshold, for the respective unit. If the output is less than the threshold, the successive testing of other units in the series is continued. If the output reaches its threshold, a gain reduction procedure is initiated on a DSP unit in a negative feedback manner. The multistage gain reduction procedure may be carried out on the unit under test or a unit upstream of the DSP under test. The gain reduction may performed internal to the DSP or by using an attenuation unit 3 at an input of the DSP The gain reduction may be a function of an amount by which the output exceeds the threshold. Modules 2 may take one or more data samples of the output and select or derive a value for testing, the value being a maximum, aggregate, or average value. Multiple DSPs may be sampled simultaneously but tested in sequence, and a specific DSP may be selected for gain reduction.

FIGURE 2

1/4

2/4

3/4 ο

ΓΝ

4/4

Application No. GB 1706797.6

RTM

Date :7 March 2018

Intellectual

Property

Office

The following terms are registered trade marks and should be read as such wherever they occur in this document:

MatLab

Intellectual Property Office is an operating name of the Patent Office www.gov.uk/ipo

DIGITAL SIGNAL PROCESSING

The present disclosure relates to methods and apparatus for avoiding or reducing signal clipping in interconnected digital signal processing units.

Digital signal processing (DSP) is widely applied in many fields dealing with signals, for example radio signal analysis, engine control systems, satellite image processing, geological analysis and electrocardiography. In audio signal processing, it is rapidly replacing its traditional, analogue counterpart. This is mainly due to cheaper and faster hardware chips as well as advances in digital audio programming, which make it possible to achieve similar sounding and even better audio effects. For example, the bass and treble knobs on an analogue radio can easily be implemented using digital filters in C-code or dedicated DSP chip assembly language.

All signals contain noise, and a goal when processing any kind of signal is to minimize noise in order to get a better signal. Weak signals generally contain more noise due to unavoidable background noise. Amplifying the signal mitigates this. However, if a signal is amplified too much, peaks of the signal might be larger than the acceptable range of the signal processing units involved, which leads to signal clipping. This introduces unwanted distortion, which alters and damages the signal irreversibly. On typical mixer panels used in music production, the input signal gain is· adjusted with a gain knob. Visual meters indicate the peak levels and the sound engineer manually adjusts the knob to achieve good signal strength while avoiding clipping.

In digital signal processing units using integer numbers, the acceptable range is fixed, e.g. 16-bit integer values between -32768 and + 32767. Any larger number will be forced down to this interval and the signal gets clipped. One way around this is to use floating-point numbers capable of representing very large and very small numbers. In this situation, clipping will no longer occur, since the acceptable range is all but infinite. However, not all signal processing works well in such a context In particular, non-linear processing such as wave shaping or compression needs a well-defined interval for optimal operation. A signal that is too strong will not match the desired processing and may give a bad result. Therefore, signals may still need to be constrained, even when using floating-point numbers.

A problem arises when connecting multiple digital signal processing units in a chain, e.g. an electric guitar connected to a number of effect units and then to a guitar amplifier simulator unit. The output level of each unit can reach levels that are too high, which will result in clipping either immediately or further down the chain. The sound engineer will hear the unwanted distortion, and will need to discover which unit in the chain needs adjustment through trial and error. This can be both distracting and time consuming. Adjusting the wrong unit might alter the sound in undesired ways and could also introduce noise by attenuating the signal too much.

Some signal processing units may incorporate anti-clipping or limiting techniques. For example, the input stage of a mixer may have a limiter to protect the equipment from peaks and to gracefully reduce amplitude without causing too much distortion. While offering protection, it will colour the sound and reduce the dynamic range of the signal.

Furthermore, consider a chain of effects where clipping is introduced early, e.g. upstream in a series or chain of digital signal processing units. Many of the downstream units might experience this clipping and, if not coordinated, decide to adjust their volumes simultaneously. This is not optimal, since only a single unit needs adjustment. As mentioned previously, this may lead to bigger unwanted changes in sound than necessary.

It is an object ofthe present invention to provide improvements in managing signal clipping in interconnected digital signal processing units.

According to one aspect, the present invention provides an automated adaptive anticlipping method for use with a series of digital signal processing units, the method comprising:

testing each one of the DSP units for an output that reaches or exceeds a predetermined threshold for the respective DSP unit;

if the output of a DSP unit under test is less than the respective threshold, continuing to successively test other units in the series;

if the output of the DSP unit under test reaches its respective threshold, initiating a gain reduction procedure on at least one DSP unit in the series.

The gain reduction procedure may be carried out on the DSP unit under test. The testing of each one of the DSP units may comprise taking one or more successive data samples of the output and selecting or deriving a value for test based on those data samples. The value for test may comprise one of: a selected maximum sample; an aggregate value; an average value; one or more values within a percentile range; a subset of a plurality of samples. The gain reduction procedure may be performed by the DSP unit under test using a gain reduction procedure internal to the DSP unit. The gain reduction procedure may comprise introducing or adjusting an attenuation unit at an input of the DSP unit under test. The gain reduction procedure may comprise reducing a gain of a DSP unit upstream of the DSP unit under test. The gain reduction procedure may comprise applying a gain reduction which is a function of an amount by which a tested output of the DSP unit exceeds the predetermined threshold. The gain reduction function may be one of a proportional function, an exponential function, a logarithmic function or a stepwise function. The method may include periodically repeating the method for the series of DSP units. The method may include resetting a gain of one or more of the DSP units. The predetermined threshold may equal a clipping threshold for the respective DSP unit under test. The successive testing each one of the DSP units may be aborted and restarted if a gain reduction procedure is initiated. Multiple ones of the DSP units may be sampled simultaneously but tested in sequence. Multiple ones of the DSP units may be tested and a specific DSP unit selected for the gain reduction procedure according to one or more predetermined criteria.

The method may comprise the steps:

if the output of all DSP units under test are less than their respective thresholds, continuing to periodically test the DSP units;

if the output of one or more of the DSP units under test reaches its respective threshold, initiating the gain reduction procedure on at least one of the DSP units under test.

According to another aspect, the invention provides an automated adaptive anti-clipping device configured to carry out any of the methods as defined above.

According to another aspect, the invention provides a digital signal processing system comprising a series of digital signal processing units and an automated adaptive anticlipping device as defined in the preceding paragraph.

Embodiments of the present invention will now be described by way of example and with reference to the accompanying drawings in which:

Figure 1 shows a schematic functional block diagram of a digital signal processing system incorporating an adaptive system for controlling signal clipping;

Figure 2 shows a schematic functional block diagram of an alternative digital signal processing system incorporating an adaptive system for controlling signal clipping;

Figure 3 shows a flow diagram of a clipping control algorithm implemented by the control module of figure 1 or figure 2;

Figure 4 shows an alternative flow diagram of a clipping control algorithm implemented by the control module of figure 1 or figure 2.

With reference to figure 1, a series of digital signal processing (DSP) units 1 (DSP1, DSP2, DSP3, DSP4) are interconnected. Each DSP unit 1 receives a respective input h, b, I3,14 and performs a signal processing function thereon to provide a respective output Οι, O2, O3, O4 which can serve as the input to a downstream DSP unit 1. In an audio system, each of the DSP units 1 may be configured for a particular audio processing function, e.g. each serving as one or more of numerous different types of sound effect generator, including amplifiers, equalisers, mixers, filters, signal addition and subtraction units, reverb or echo units, modulators, compressors, synthesisers, noise cancellers, overdrive effect units, etc.

The output Οι, O2, O3, O4 of each DSP unit 1 can be respectively sampled by sampling modules 2 (S1, S2, S3, S4). The samples are fed to a clipping control module 5 for the detection of output levels of each module, according to an algorithm as discussed below. The clipping control module 5 is configured to control an output gain level of each DSP unit, e.g. by control lines 6, 7, 8, 9.

The number and interconnection arrangement of DSP units 1 is illustrative and can be varied according to requirement, e.g. by the addition of further DSP units either fully in series as illustrated in figure 1, or possibly including branching interconnections, such as where further input streams are combined in selected DSP units 1 or separated for subsequent parallel processing in splitters, for example.

The DSP units 1, sampling modules 2 and clipping control module 5 may, for example, be implemented as individual electronic processing modules (e.g. implemented in hardware, firmware or software or a combination thereof) or may be implemented as one or more processes or threads running on a generic computer system / electronic processing device or on an application-specific computer system. In one example, the clipping control module 5 may be implemented as a parallel process thread which executes simultaneously with one or more process threads implementing the functions of the DSP units 1.

Figure 2 illustrates an alternative arrangement of digital signal processing units 1 similar to figure 1 but in which upstream of each input h, b, b, I4 of each DSP unit 1 is an attenuation module 3 for providing a reduced level of input to the respective DSP unit 1. As will become apparent in the further discussion below, the attenuation units 3 may not need to be provided for each and every DSP unit 1 but might be provided only for selected ones.

Figure 3 illustrates an algorithm 20 for controlling signal clipping implemented by the clipping control module 5 of figure 1 or figure 2. The signal clipping control module 5 periodically samples the outputs of each of the DSP units 1 (box 21). The control module tests each sampled output to see if it reaches or exceeds a predetermined threshold for that respective DSP unit (box 22). If the DSP unit output under test is less than the threshold, the control module 5 moves on to another DSP unit in the system (box 23) and continues to successively test other units (box 21). If the output of the DSP unit under test reaches its respective threshold, the control module 5 initiates a gain reduction procedure (box 24).

In the context of the example of figure 1, the gain reduction procedure 24 may comprise triggering a gain reduction procedure that is internal to the respective DSP unit 1 (box 25). This may include driving a gain control function to a new setting or initiating an automatic gain control procedure within the DSP unit 1.

Alternatively, in the context of the example of figure 2, the gain reduction procedure 24 may comprise adjusting an attenuation module 3 at the input of the DSP 1 (box 26). It will be understood that this adjusting function may comprise introducing an attenuation function module (e.g. a process thread) between the respective DSP units 1, or adjusting the attenuation factor being applied in an existing attenuation module / process thread.

After initiating a gain reduction procedure, the clipping control module 5 may restart the process by returning to test the first DSP unit in the sequence of DSP units (box 27).

Alternatively, it may continue to test another DSP unit in the sequence of DSP units. In this process, the next DSP unit could be the next DSP unit downstream of the DSP unit just tested, or could be a return to the first DSP unit in the sequence or could be a different DSP unit in the sequence.

In one example process, the DSP units are scanned by the control module 5 periodically. One example could be to scan the DSP unit outputs, e.g. every 100 ms. The frequency of iterating the routine exemplified by figure 3 can be varied according to the application. In audio signal processing, the period of 100 ms offers benefits for quick control of clipping and distortion without overburdening a processor which is multitasking between the signal processing thread and the clipping control thread. If the clipping control algorithm is operating as an independent process thread, the scan could be implemented in real time, e.g. by using data samples as they pass through the DSP chain. Thus, the period between each scan could be less than 100 ms or more than 100 ms. The period between clipping control scans could be determined according to a number of data samples processed in the DSP unit chain, rather than a strict time period.

The testing of the DSP unit outputs (box 22) can be carried out on the basis of individual samples or aggregated samples, as captured by the sampling modules 2 (box 21). For example, the testing against a target threshold (box 22) may be carried out on a single data sample or on a plurality of successive samples. The testing may be carried out on an average of a number of samples, or on a maximum value sample in a succession of samples, or on a number of samples meeting some criteria such as the highest n samples in a set of samples >n, e.g. the samples within a certain percentile level. The samples may be averaged, and the sample set could have outliers or extreme values removed before testing. Numerous other strategies could be deployed to ensure that the testing (box 22) is performed on data that is optimised to achieve reliable and optimally performing operation of the system. Thus, in a general aspect, the process may select or derive a value for testing against the threshold based a set of successive data samples.

In a preferred arrangement, each iteration of the process flow 20 tests the peak level of the output of each DSP unit 1 in turn, starting with the first DSP unit DSP1 and progressing through the chain, e.g. to DSP4.

The target threshold (box 22) is, in one example, defined as the clipping point (0 dBFS,

i.e. +/-1). If the absolute peak level is below the threshold, the process 22 moves on to the next DSP unit, DSP2 (box 23). Once the last unit DSP4 has been tested and if no clipping was detected, the sample values at the sampling modules can be reset and the process returns to the start. The target threshold can be defined in a number of different ways, including at a level which is somewhat below a clipping level, e.g. as a percentage of the clipping level to provide some headroom before clipping commences.

If, at step 22, the absolute peak level is equal to, or higher than the threshold, the DSP unit whose output is being tested is adjusted using the gain reduction procedure (box 24). In the figure 1 arrangement, the adjustment seeks to reduce the internal gain ofthe DSP unit 1. How this is best achieved can be specific to the type of DSP unit 1 being tested. The DSP unit itself may therefore best solve it, by exposing a well-defined, generic interface to the clipping control module 5 for this operation. The gain adjustment may be performed to be a gain reduction which is a function of an amount by which the tested output ofthe DSP unit exceeds the predetermined threshold. The function may be a proportional function, an exponential function, a logarithmic function or a step-wise function. In other words, the gain reduction factor may for example, be proportional to the extent to which the sampled and tested output exceeds the threshold. In the figure 2 arrangement, the adjustment of the attenuation module may be configured to achieve a similar level reduction to the output of the respective DSP unit 1 according to the same principles, e.g. as a function of the amount by which the tested output of the DSP unit exceeds the predetermined threshold.

If a gain or attenuation adjustment is made, the process can be terminated for this scan and the procedure 20 is initiated again after a suitable time interval. This is preferred where clipping peaks further downstream in the signal path might be false positives due to the overdriving ofthe DSP unit just identified for which a gain reduction is made.

Alternatively, if the sampling is being performed on a real time basis such that a new set of sample data is available for the testing ofthe subsequent DSP unit, it may be appropriate to continue with testing the next DSP unit in the sequence.

Other techniques for gain reduction or attenuation of a DSP unit output can be considered. If a DSP unit 1 does not provide a way to adjust its internal gain, the additional attenuation modules 3 of figure 2 may be deployed, or a gain adjustment may be made to a DSP unit upstream ofthe DSP unit under test, and preferably ofthe DSP unit immediately preceding the DSP unit under test. This can lead to the desired result, depending on how the DSP unit works internally. Care is taken to ensure that any nonlinearities which create differences in how input and output levels correspond to each other are accommodated so that the adjustment is not too big or too small, and that there will be no other side effects which could produce unwanted changes in sound.

For selected DSP units, where adequate control of gain or attenuation at the input is not feasible, these units might be ignored and the next DSP unit downstream tested. If that downstream unit also shows high peak levels, it can be adjusted instead. While less optimal, there is still a reasonable chance that the end result is good, especially if both units can handle high peaks without noticeable problems or clipping. Many floating-point based linear processing unit implementations conform to this requirement, e.g. infinite impulse response filters and delays.

The shorter the scan interval (i.e. the time interval before starting a new set of tests on the DSP units according to process 20), the more clipping is avoided. For example, in one arrangement, the DSP units may be configured to process between 128 and 256 samples in 3 millisecond blocks before each new clipping test process 20.

Clipping can be more effectively avoided also by setting the target threshold (box 22) somewhat below the clipping level of 0 dBFS, particularly if combined with very short testing intervals.

While a sound engineer might be able to identify and reduce clipping manually, the system described herein has a number of advantages. It is very fast and responsive and can remove clipping within 100 milliseconds or less. It can adjust the correct or most appropriate DSP unit 1 directly and avoid unnecessarily adjusting other DSP units in the system. The perceived sound is thereby preserved as much as possible. The system can be tuned to perform exactly the right amount of gain reduction or attenuation to avoid clipping. This also helps preserve the perceived sound. The system enables the sound engineer to concentrate on the intended task of sound production without being distracted by the necessity of making continual adjustments to different DSP units.

Compressors and limiters are frequently used to avoid clipping. This algorithm has the advantage of preserving the original signal dynamics, leaving the signal less affected, less distorted and therefore perceived as less coloured. As an example, an overly amplified signal sent through a compressor / limiter may be severely limited and lose most of its dynamic range. A listener may perceive it as being very coloured. The process described herein preserves the dynamic range, and the listener may perceive it as completely unaffected and uncoloured.

In addition to avoiding clipping, the process may help to maintain average signal levels within a more confined range. This reduces the need for adjusting levels e.g. when mixing music. This is especially desirable when combining many sound sources such as multitrack mixing, where the different sources need to be in balance with each other.

The algorithm is applicable to many different audio processing applications. It can work on any digital signal processing unit capable of adjusting its internal gain, without any knowledge of the inner workings of the unit. This generic nature makes it suitable for largescale use in host / plugin architecture, where many independent unit developers can produce digital signal processing units as plugins for a host environment. The host environment can then be responsible for connecting the units and for executing the clipping control process across all plugins.

While the illustrative examples herein describe the application to audio signal processing, It will be understood that the process is applicable to any type of signal processing chain where clipping needs to be avoided. For instance, it could be useful in the MatLab programming environment where scientists are able to model signal processing. Matlab is used across all disciplines, e.g. radio analysis.

A preferred operation of the process flow 20 is to test each DSP unit 1 in turn, i.e. to follow steps 21-23 repeatedly for each successive DSP unit in the chain of DSP units, DSP1 ... DSP4. If a DSP unit is found to be at or above the threshold then the process branches into steps 24-26 for that DSP thereby completing the process and waiting for the predetermined time interval before commencing a new process at DSP1. However, it will be understood that the process is not constrained to test each DSP unit in succession of their positions in the process sequence.

For example, key DSP units that are most susceptible to clipping could be prioritised in the testing sequence. In another arrangement, each DSP unit could be sampled and / or tested substantially simultaneously and a decision on which DSP unit should be selected for the gain reduction procedure made by the clipping control module 5 according to another criterion or criteria. Such a criterion could include selecting the DSP unit exceeding its respective threshold by the greatest amount, or selecting a DSP unit furthest upstream in the signal processing chain or by reference to a priority table. An example of such a process variation is shown in figure 4.

With reference to figure 4, an algorithm 30 is illustrated in which the clipping control module 5 of figure 1 or figure 2 periodically samples the outputs of each of the DSP units 1 (box 31) in series or in parallel. The control module tests each sampled output, in series or in parallel, to see if it reaches or exceeds a predetermined threshold for the respective DSP unit (box 32). If none of the DSP unit outputs under test reach or exceed the target threshold (box 33), the control module 5 repeats the testing periodically, e.g. after a specified delay (box 34). If any of the outputs of the DSP units under test reach their respective thresholds, the control module 5 effects a selection process for determining which unit or units will be targeted for a gain reduction procedure (box 35). The selection process could comprise one or more criteria. These criteria could include, for example: which DSP unit has exceeded its target by the largest amount; which DSP unit is most susceptible to provide a highly distorted or corrupted output when clipping or which is most likely to result in noticeable output degradation; which DSP output unit has the largest effect on the output of the series of DSP units; etc. The control module 5 then initiates a gain reduction procedure according to the selection criteria (box 35).

Again in the context of the example of figure 1, the gain reduction procedure 35 may comprise triggering a gain reduction procedure that is internal to the respective DSP unit 1 (box 36), as previously described. Alternatively, in the context of the example of figure 2, the gain reduction procedure 35 may comprise adjusting an attenuation module 3 at the input of the DSP 1 (box 37), as previously discussed. This adjusting function may comprise introducing an attenuation function module (e.g. a process thread) between the respective DSP units 1, or adjusting the attenuation factor being applied in an existing attenuation module / process thread.

After initiating a gain reduction procedure, the clipping control module 5 may restart the process by returning to test the DSP units (box 38).

Thus, in a general aspect, it can be seen that the method encompasses an automated adaptive anti-clipping method for use with a series of digital signal processing units which comprises: (i) testing each of the DSP units for an output that reaches or exceeds a predetermined threshold for the respective DSP unit; (ii) if the output of all DSP units under test are less than their respective thresholds, continuing to periodically test the DSP units; and (iii) if the output of one or more of the DSP units under test reaches its respective threshold, initiating a gain reduction procedure on at least one of the DSP units under test. In another general aspect, it can be seen that the method encompasses an automated adaptive anti-clipping method for use with a series of digital signal processing units which comprises: (i) testing each of the DSP units for an output that reaches or exceeds a predetermined threshold for the respective DSP unit; and (ii) if the output of one or more of the DSP units under test reaches its respective threshold, initiating a gain reduction procedure on at least one of the DSP units under test.

Various strategies may be deployed for restoring signal levels after a clipping event has forced a gain reduction or attenuation control. For example, an automatic gain control in a DSP unit may operate to gradually restore gain to that unit at some time after a forced reduction by the clipping control module 5. An automated attenuation control may be configured to gradually restore the attenuation factor back to unity in the attenuation modules 3 at some time after a forced reduction. It may be left to a user to manually restore gain at some point in the signal processing.

Performing gain reduction as a function of an amount by which a tested output of a DSP unit exceeds a threshold might not be available if the threshold is set at a clipping threshold, and may operate best when there is still headroom for the signal to exceed the threshold.

Other embodiments are intentionally within the scope of the accompanying claims.

Claims

1. An automated adaptive anti-clipping method for use with a series of digital signal processing units, the method comprising:

2. The method of claim 1 in which the gain reduction procedure is carried out on the DSP unit under test.

3. The method of claim 1 in which the testing of each one of the DSP units comprises taking one or more successive data samples of the output and selecting or deriving a value for test based on those data samples.

4. The method of claim 3 in which the value for test comprises one of: a selected maximum sample; an aggregate value; an average value; one or more values within a percentile range; a subset of a plurality of samples.

5. The method of claim 1 in which the gain reduction procedure is performed by the DSP unit under test using a gain reduction procedure internal to the DSP unit.

6. The method of claim 1 in which the gain reduction procedure comprises introducing or adjusting an attenuation unit at an input of the DSP unit under test.

7. The method of claim 1 in which the gain reduction procedure comprises reducing a gain of a DSP unit upstream of the DSP unit under test.

8. The method of any one of claims 5, 6 and 7 in which the gain reduction procedure applies a gain reduction which is a function of an amount by which a tested output of the DSP unit exceeds the predetermined threshold.

9. The method of claim 8 in which the gain reduction function is one of a proportional function, an exponential function, a logarithmic function or a stepwise function.

10. The method of claim 1 further including periodically repeating the method for the series of DSP units.

11. The method of claim 8 further including resetting a gain of one or more of the DSP units.

12. The method of claim 1 in which the predetermined threshold equals a clipping threshold for the respective DSP unit under test.

13. The method of claim 1 in which the step of successively testing each one of the DSP units is aborted and restarted if a gain reduction procedure is initiated.

14. The method of claim 1 in which multiple DSP units are sampled simultaneously but tested in sequence.

15. The method of claim 1 in which multiple DSP units are tested and a specific DSP unit is selected for the gain reduction procedure according to one or more predetermined criteria.

16. The method of claim 1 in which:

17. An automated adaptive anti-clipping device configured to carry out the method of any one of claims 1 to 16.

18. A digital signal processing system comprising a series of digital signal processing units and an automated adaptive anti-clipping device according to claim 17.

Intellectual

Property

Office

Application No: GB1706797.6 Examiner: Dr Laurence Drummond Claims searched: 1-18 Date of search: 5 March 2018