Dynamic digital IIR audio filter and method which provides dynamic digital filtering for audio signals

Abstract

An improved dynamic digital IIR (Infinite Impulse Response) audio filter and method provides a remedy for the major problems in state of the art digital IIR filtering and dynamic digital filtering for audio. In particular, the required multiplier coefficient size is reduced, such that 16 bit coefficients are adequate for good frequency resolution. In addition, the data size of the multiplier is also reduced, to be similar to the input and output data resolution without inferior noise performance. Furthermore, the structure is easily modified to allow the fixing of the DC (Direct Current) gain of the IIR filter to unity regardless of coefficient choice, eliminating any problems with filter dynamic range at low frequency passbands. Additionally, a dynamic filter structure allows logarithmic filter sweeps and compresses the coefficients further down to as few as 8 bits with adequate resolution for audio throughout the band. The structure allows the alteration of sweep of the filter without excessive computation or the requirement to read the current sweep value.

Description

BACKGROUND OF THE INVENTION

The present invention relates to a dynamic digital IIR audio filter and corresponding method.

The current state of the art for digital infinite impulse response (IIR) filters for audio applications utilizes long wordlengths to obtain adequate noise performance for adequate sound quality. Typically 24 bit or longer coefficients are used to obtain adequate resolution at low frequencies, and 24 bit multiplier data paths or multiple precision techniques are used to improve noise performance. Finally, extreme dynamic range problems exist when filters with very low frequency passbands are used. The article in the Jan. 1989 Journal of the Audio Engineering Society provides a summary of the current state of the art.

The state of the art in dynamic audio filtering, in which the coefficients are to be changing in time, is even more primitive. Coefficients are typically linearly interpolated, producing filters whose characteristics typically vary dramatically in audio quality over the sweep range from the desired swept response. The coefficients are linearly encoded in the classical IIR filter forms, preventing any logarithmic or audibly meaningful sweep, and requiring long words of typically 24 bits for adequate low frequency resolution. Furthermore, any dynamic sweeping requires computation of the increment required to be added each sample time. This complicates control of the filter during complicated sweeps, and makes it impossible to change the sweep trajectory of the filter mid-sweep without reading back the current sweep location.

SUMMARY OF THE INVENTION

It is an object of the present invention to provide an improved dynamic digital IIR audio filter and method.

It is another object of the present invention to remedy the major problems in state of the art digital IIR filtering and dynamic digital filtering for audio. In particular, the required multiplier coefficient size is reduced, such that 16-bit coefficients are adequate for good frequency resolution. In addition, the data size of the multiplier is also reduced to be similar to the input and output data resolution without inferior noise performance. Furthermore, the structure is easily modified to allow the fixing of the DC gain of the IIR filter to unity regardless of coefficient choice, eliminating any problems with filter dynamic range at low frequency passbands.

Additionally, a dynamic filter structure is disclosed which allows logarithmic filter sweeps, and compresses the coefficients further down to as few as 8 bits with adequate resolution for audio throughout the band.

Finally, a structure is disclosed which allows the alteration of sweep of the filter without excessive computation or the requirement to read the current sweep value.

All of these objectives can be accomplished by the present invention while allowing implementation of up to 128th order filtering on a single silicon integrated circuit.

Additional objects, advantages and novel features of the present invention will be set forth in part in the description which follows and in part become apparent to those skilled in the art upon examination of the following or ay be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations which are pointed out in the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings which are incorporated in and form a part of this specification illustrate an embodiment of the invention and, together with the description, serve to explain the principles of the invention.

FIG. 1 depicts a diagram of a prior art digital filter structure.

FIG. 2 depicts another diagram of a prior art digital filter structure.

FIG. 3 depicts a diagram of an improved digital filter structure according to the present invention.

FIG. 4 depicts a diagram of another embodiment of a digital filter structure according to the present invention.

FIG. 5 depicts a further feature of the digital filter according to the present invention.

FIG. 6 depicts an implementation of the improved dynamic digital audio filter according to the present invention.

FIG. 7 depicts a diagram of a multiplier block which forms a portion of FIG. 6.

FIG. 8 depicts a diagram of a decoding technique which forms a part of the present invention.

FIG. 9 depicts a preferred embodiment of the present invention.

FIG. 10 depicts logic used to update an interpolating variable.

FIG. 11 illustrates circuitry used to implement automatic computation of an encoded increment according to the present invention.

FIGS. 12A and 12B show the logic diagram of a portion of FIG. 11

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Reference will now be made in detail to the preferred embodiment of the invention, an example of which is illustrated in the accompanying drawings. While the invention will be described in conjunction with the preferred embodiment, it will be understood that it is not intended to limit the invention to that embodiment. On the contrary, it is intended to cover alternatives, modifications and equivalents as may be included within the spirit and scope of the invention as defined by the appended claims.

The classical filter equation for a second order IIR digital filter is

Yn=Xn-B1*Yn-1-B2*Yn-2 B1 and B2 form complex conjugate poles at locations: R=sqrt(B2) theta=arccos(-B1/2R) where R is the radius of the pole location and theta is the angle, with zero coresponding to zero frequency and pi coresponding to half the sampling rate or approximately 20 kHz. A classical filter structure to directly implement this equation is shown in FIG. 1. Additionally, a second order FIR stage can be added to produce zeros to complete the classical structure as shown in FIG. 2.

The audio spectrum spans frequencies from 20 Hz to 20 kHz, and is typically logarithmically divided into ten octaves by the ear. The ear's sensitivity to loudness is similarly logarithmic and is approximately a range of 100 decibels (dB), where a decibel is defined as

20* log10(E/Eref) where E is the signal level and Eref is a reference level.

It is easily seen that the magnitude response of a resonance in a filter produced by a pole is proportional to the difference between the pole's radius and unity. Hence the ear's sensitivity to the position of the pole is tremendously amplified as the pole's radius approaches unity. Hence it would be desirable to develop a filter whose coefficients were proportional to the distance between the pole's radius and unity.

Through a similar argument, emphasizing instead the ear's logarithmic frequency sensitivity, it can be shown that the desirable coefficient relationship for the angle of poles should be such that small numbers represent very small angles, whose cosines approach unity.

If one examines the flow graph of an IIR filter from a noise point of view, one will observe that the quantization of the signal at the input to the multiplier is the source of quantization noise. Reduction in the required truncation at the multiplier input will reduce the noise produced by the filter. It is easily observed that IIR filters become noisier as their poles tend towards unit radius and towards very low or very high frequencies. This can be explained by realizing that in these cases, the recursion within the filter is more severe, hence increasing the adverse effect of the multiplier input truncation noise. If the filter structure can be altered such that in these cases there is less effective quantization noise, improved performance in these interesting cases will result. A filter structure which accomplishes the above structural requirements according to the present invention is shown in FIG. 3, and the extension to add FIR terms in FIG. 4. Note that the coefficients have been modified as per the requirements:

B1'=B1+2 B2'=1-B2 and hence all coefficients are always positive, and the multiplies unsigned.

Another desirable feature for audio filters, in keeping with the objects of the invention, is to force the gain of the filter at zero frequency to be unity. This feature can be accomplished by the structure in FIG. 5. Note that the structures in FIGS. 4 and 5 are easily switched between.

One implementation of the filter according to the present invention is shown in FIG. 6. In one preferred embodiment, a total of four multiplies and eight additions are used. As will later be discussed, this allows four multiplies to be used for interpolation of coefficients in dynamic filtering if the adder and multiplier have the same speed, which is typically the case using modern VLSI components.

In FIG. 6, accumulator 10 is used to accumulate the output signal. Auxilliary accumulator 11 accumulates the value to be saved in the delay memory 22. The filter operates in eight states. At the beginning of the cycle, accumulator 10 is loaded with the input signal.

During the first cycle, the adder 12 is supplied with negative delay memory 2's data from latch 13 at input 14 via multiplexer 15, and at input 17 the input signal from accumulator 10, or zero, depending on the DC gain stabilization desired, through multiplexer 16. The sum is latched into the multiplier input latch 18 to be multiplied by the associated unsigned coefficient B2' from coefficient memory 23, the output of which will simultaneously be loaded into multiplier input latch 19.

During the next cycle, negative delay memory 1's data from latch 20 is similarly routed to adder input 14, while input 17 stays the same. The sum is latched again in multiplier input latch 18 for multiplication by unsigned coefficient B1'.

During the third cycle, adder input 14 is switched back to negative delay memory 2's data, while adder input 17 is changed by multiplexer 16 to signal 21, which is positive delay memory 1's value shifted left one bit to form twice delay memory 1's value. The resulting sum is stored at the end of the cycle in auxilliary accumulator 11.

During the fourth cycle, the output of the multiplier 24 is routed to added input 14. This now is the product resulting from coefficient B2'. Adder input 17 will be selected as either negative the input signal as routed from accumulator 10, or zero, depending on the DC gain stabilitzation requirement. The resulting sum is latched into accumulator 10.

During the fifth cycle, the output of the multiplier 24, containing the product resulting from coefficient B1', is routed to adder input 10. The negative of the contents of accumulator 10 is routed to adder input 17. The resulting sum is stored in accumulator 10.

During the sixth cycle, the contents of accumulator 11 is routed to adder input 14, and of accumulator 10 to input 17. The resulting sum is saved in accumulator 11 to be stored in delay memory 22.

During the seventh cycle, the multiplier output, which is the product resulting from coefficient A2', is routed to adder input 14, and the negative of accumulator 10 routed to input 17. The resulting sum is stored in accumulator 10.

During the eighth and final cycle, the multiplier output, which is the product resulting from coefficient A1', is routed to the adder input 14, and the negative of the signal accumulator is routed to input 17. The resulting sum, which is the filter output and hence the input to the next stage, is stored in accumulator 10.

The reservation table for the above pipeline described in conjunction with FIG. 6 is presented below.

RESERVATION TABLE__________________________________________________________________________ Adder A 10 A 11Adder Input 14 Adder Input 17 CD Result contents contents Prod__________________________________________________________________________Z.sub.2 ' (SA = X.sub.n) & DC 1 DC & X.sub.n - Z.sub.2 X.sub.n xx xxZ.sub.1 ' (SA = X.sub.n) & DC 1 DC & X.sub.n - Z.sub.1 X.sub.n xx xxZ.sub.2 ' 2Z.sub.1 1 2Z.sub.1 - Z.sub.2 X.sub.n xx xxb.sub.2 '(DC & X.sub.n - Z.sub.2) (SA = X.sub.n ') & IDC IDC -AR2 X.sub.n 2Z.sub.1 - Z.sub.2 b2(DC & X.sub.n - X.sub.2)b.sub.1 '(DC & X.sub.n - Z.sub.1) SA = -AR2' 1 b.sub.1 '(DC & X.sub.n - Z.sub.1) -AR22 2Z.sub.1 - Z.sub.2 b.sub.1 (DC & X.sub.n - Z.sub.1)2Z.sub.1 - ZX.sub.2 SA = AR 0 Y.sub.n AR 2Z.sub.1 - Z.sub.2 a.sub.2 'Z.sub.2a.sub.2 'Z.sub.2 SA' = AR' 1 a.sub.2 'Z.sub.2 - AR AR Y.sub.n a.sub.1 'Z.sub.1a.sub.1 'Z.sub.1 SA' = -MA1' 1 MA1 + a.sub.1 'Z.sub.1 -MA1 Y.sub.n xx MA.sub.out = X.sub.n__________________________________________________________________________ AR2 = IDC & X.sub.n - b.sub.2 '(DC & X.sub.n - X.sub.2) AR = IDC'X.sub.n + b.sub.1 '(DC & X.sub.n - Z.sub.1) + b.sub.2 '(Z.sub.2 DC & X.sub.n =AR2 + b.sub.1 '(DC & X.sub.n - Z.sub.1) Y.sub.n = IDC'X.sub.n - b.sub.1 '(Z.sub. 1 - DC & X.sub.n) + b.sub.2 '(Z.sub.2 - DC & X.sub.n) + .sub.2 Z.sub.1 - 2Z.sub.2 = AR + .sub.2 Z.sub.1 - 2Z.sub.2 MA1 = IDC'X.sub.n - b.sub.1 '(Z.sub.2 - DC & X.sub.n) + b.sub.2 '(Z.sub.2 - DC & X.sub.n) - a.sub.2 'Z.sub.2 = AR - a.sub.2 MA.sub.out = IDC'X.sub.n - b.sub.1 '(Z.sub.1 - DC & X.sub.n) + b.sub.2 '(Z.sub.2 - DC & X.sub.n) + a.sub.1 'Z.sub.1 - a.sub.2 'Z.sub.2 = MA1 + a.sub.1 'Z.sub.1 Notes: Y.sub.n must be positive, hence XL must be positive 2Z.sub.1 - Z.sub.2 or dodi must be complemented, dodi cannot always be complemented as both senses of product area required for MA portion and we cannot add two complements as a carry in of 2 is required. Hence, Z.sub.2 must be complemented, therefore Z.sub.1 should be complemented. We conclude that the B multiplies will be negative Z data.

It should be noted that the multiplier 24 of FIG. 6 uses block floating point techniques on the coefficients to achieve a greater range. The details of the multiplier block are illustrated in FIG. 8. The signal is applied to the signed input 31 of the multiplier input stage 32 (which consists of the Booth's recoders and wallace trees), and the coefficient mantissa is applied to the unsigned input 33. The coefficient exponent is applied to the exponent pipeline latch 34, which is required due to the multiplier pipeline latch 35 which feeds the second stage 36 of the multiplier consisting of the final lookahead carry adder. This in turn feeds barrel shifter 37, which adjusts for the previous coefficient's exponent. The final product, correctly scaled, is stored in product latch 38.

In accordance with the objects of the current invention, logarithmic interpolation of the coefficients is required to produce audibly meaningful sweeps. This is accomplished not through true logarithmic interpolation, but rather through linear interpolation of approximately logarithmically encoded coefficients.

The poles of the filter are assumed to be, for important cases, near zero frequency and near unity radius. Hence, B2'=1-B2=1-R2=1-(1-2d+d2) if R=1-d. Since R is close to 1, d is small and d2 is very small, so B2' is approximately 2d, and hence proportional to the distance from the pole to the unit circle. Consequently a nominally logarithmic coding for B2' will be an audibly meaningful coding.

Similarly, if the poles angle t is near zero, cos(t) will be approximately 1-t2, hence B1'=2+B1=2-2Rcos(t)=2-2(1-d)(1-t2)=2d+2(1-d)t2 which is approximately B2'+2t2. One can thus logarithmically encode the difference between B1' and B2' and accomplish the objective of approximating logarithmic frequency and resonance control.

The actual decoding can be accomplished by means of either a ROM or a shifting technique. The latter is demonstrated in FIG. 9.

In FIG. 9, the coefficient to be decoded is held in coefficient latch 41. In the case of a static filter, this would be supplied from a memory holding the coefficients; in the case of a dynamic filter, this would be the output of the coefficient math unit, which would add the lower coefficient to the product of the interpolating variable times the coefficient differences. The most significant bits 42, (in this case 4 bits are used), are routed through the shift decoding logic 43, to determine the degree of shift in the barrel shifter 44. The resulting decoded coefficient is stored in latch 45. If block floating point coefficients are used, the shift logic can additionally or alternatively output an exponent into exponent latch 46. If the coefficient is a second coefficient (B2' or A2'), the result is directly passed to the final output latch 47 through multiplexer 48, and exponent control logic 49 passes the exponent unchanged into exponent output latch 50 if block floating point is used.

If the coefficient is a first coefficient (which must be processed second), then adder 51 adds the previously computed second coefficient as stored in 47 to the decoded difference in latch 45, and the result passes through multiplexer 48 into the output latch. If block floating point is used, the exponent logic 49 must control barrel shifters 52 and 53 to normalize the addition.

The coefficients are interpolated according to the formula C(x)=Ca+x(Cb-Ca), where x varies from zero to unity. This requires one multiply per coefficient, and with two coefficients each for poles and zeros one can use the available four multipliers to make complete use of the required hardware in a time division multiplexed system.

The interpolating variable x is updated in the known manner of incrementing its value each sample period, and clamping the result to a final known value when the clamp value is exceeded. This works well when both the increment and clamp are computed externally as the dynamic sweep rate and destination. However, this method fails when a slow controlling processor is generating the sweep variables, and a real time command requires the change to a new destination using a sweep while the current sweep is in progress. In this case, the slower processor must calculate new new sweep increment based on the current value of the interpolating variable x and the destination value. Unfortunately, the interpolating variable x continues to be incremented according to the old sweep while the slow processor calculates its result, with an inaccurate (or even worse, the wrong signed) increment saved when the computation is done.

The solution to this problem is another aspect of the present invention. Rather than supplying the interpolating hardware with both a destination and increment, the interpolator is supplied with a destination and an approach time in samples. The interpolator computes within a single sample period the difference between the destination and current value, divides this by the time, and stores the result in the increment.

The preferred embodiment of this solution will now be described. The increment is stored in a compressed value as indicated in Table I. The 8-bit value S, shift2, shift1, shift0, d3, d2, d1, d0 is translated into a signed 16-bit linear value from -20408 to 2047, using the logic circuitry shown in FIG. 10. Sign bit S and data bits d3, d2, d1 and d0 are shifted in a sign-extended matter by barrel shifter 60. The control of the shifter is performed by logic 61 to conform to the translation defined in Table I below.

TABLE I__________________________________________________________________________16bitl F d15 d14 d13 d12 d11 d10 d9 d8 d7 d6 d5 d4 d3 d2 d1 d0 shift__________________________________________________________________________7 S S S S S S* d3 d2 d1 d0 0 0 0 0 0 0 06 S S S S S S S* d3 d2 d1 d0 0 0 0 0 0 15 S S S S S S S S* d3 d2 d1 d0 0 0 0 0 24 S S S S S S S S S* d3 d2 d1 d0 0 0 0 33 S S S S S S S S S S* d3 d2 d1 d0 0 0 42 S S S S S S S S S S S* d3 d2 d1 d0 0 51 S S S S S S S S S S S S* d3 d2 d1 d0 60 S S S S S S S S S S S S d3 d2 d1 d0 6__________________________________________________________________________ NOTE: Adder carry is set for increment add during positive increments. This means that a zero increment cannot be reached (instead, set target to correct value), and that the increment ranges for positive and negative are the same (000107c1 = ffff - f83f). This simplifies automatic decoding from targetvalue, as zero cannot be reached on underflow.

The clamp value is similarly an 8-bit value, which is translated into a 16-bit linear value by simply duplicating the bits twice. For example, 5a hexidecimal would translate to 5a5a hexidecimal. FIG. 11 illustrates the logic used to add update the 16-bit interpolating variable using the 8-bit increment and clamp.

In FIG. 11, register 70 holds the current value of the 1 6-bit interpolating variable. Register 71 holds the 8-bit clamp value, and register 72 holds the 8-bit encoded increment value. Decoder 73 decodes the 8-bit increment into a 16-bit signed linear value. Signal 74 controls the two phases of the operation. In the first phase, multiplexers 75 and 76 select as the inputs to the added 77 the current value and linear increment, and multiplexer 78 is controlled by logic unit 79 so that the incremented current value appears in output register 80. Logic 79 also notes the state of the adder carry signal 81 with respect to the increment sign bit 82, such that an arithmetic overflow can be detected during this phase. In the second phase, multiplexers 75 and 76 select the inverted clamp value and incremented current value respectively, and hence adder 77 determines the difference between these values. The state of adder carry signal 81 and increment sign bit 82 now allows control logic 79 to determine if the incremented current value has exceeded the current value; that is, if the increment sign is positive and the incremented current value is greater than the clamp, or if the increment sign is negative and the incremented current value is less than the clamp. If the incremented current value exceeds the clamp, or if overlow occurred during phase 1, then logic 79 causes multiplexer 78 to select the 16-bit clamp value; otherwise the incremented current value is selected. At the end of the second phase, register 80 contains the properly computed new current value.

FIG. 12 illustrates the circuitry required to implement the automatic computation of the required encoded 8-bit increment, based on a destination (clamp) value, current value and approach time. The approach time is selected from eight possible values, each a power or two greater, with the lowest being 32 sample periods. This approach time is stored in register 90. The current value of the interpolator is stored in register 91, and the destination is stored in register 92. Adder 93 computes the difference between the destination and current value, which is then stored in register 94. Logic 95 computes the required increment to reach the destination in approach time samples. The 8-bit value is output on signal lines 96 when enable signal 97 is active.

FIG. 13 shows the detailed logic with block 95. Logic 100 determines the first set bit, that is, the first bit different from the sign bit. Logic 101 determines the shift bits of the required increment based on the first set bit and the selected approach time indicated by CUDR2N, CUDR1N and CUDRON. Signal 102 indicates that the first set bit is greater than or equal to the selected approach time, which determines the function of data bit barrel shifter 103. If the first set bit is greater than or equal to the selected approach time, the first set bit determines the shifting of the data bits. Otherwise, the selected approach time determines the shifting of the bits. Enable signal 104 simply allows the value to be driven onto the circuit outputs. When the enable is invalid, all outputs are logical 1's. Table II below illustrates the resulting outputs as a function of the difference input and the selected approach time.

TABLE II__________________________________________________________________________ARMAdillo Interpolator Increment Decoding:Difference Binary FSB @ .6 @ 1.2 @ 2.4 @ 4.8 @ 9.6 @ 20 @ 40 @ 80__________________________________________________________________________0ffff-0f800 011111xxxxxxxxxxx 15 7f 6f 5f 4f 3f 2f 1f 0f0f7ff-0f000 011110xxxxxxxxxxx 15 7e 6e 5e 4e 3e 2e 1e 0f0efff-0e800 011101xxxxxxxxxxx 15 7d 6d 5d 4d 3d 2d 1d 0e087ff-08000 010000xxxxxxxxxxx 15 70 60 50 40 30 20 10 0807fff-07c00 0011111xxxxxxxxxx 14 6f 5f 4f 3f 2f 1f 0f 0707bff-07800 0011110xxxxxxxxxx 14 6e 5e 4e 3e 2e 1e 0f 07077ff-07400 0011101xxxxxxxxxx 14 6d 5d 4d 3d 2d 1d 0e 07043ff-04000 0010000xxxxxxxxxx 14 60 50 40 30 20 10 08 0403fff-03e00 00011111xxxxxxxxx 13 5f 4f 3f 2f 1f 0f 07 0303dff-03c00 00011110xxxxxxxxx 13 5e 4e 3e 2e 1e 0f 07 03021ff-02000 00010000xxxxxxxxx 13 50 40 30 20 10 08 04 0201fff-01100 000011111xxxxxxxx 12 4f 3f 2f 1f 0f 07 03 0101eff-01e00 000011110xxxxxxxx 12 4e 3e 2e 1e 0f 07 03 01010ff-01000 000010000xxxxxxxx 12 40 30 20 10 08 04 02 0100fff-00f80 0000011111xxxxxxx 11 3f 2f 1f 0f 07 03 01 0000f7f-00f00 0000011110xxxxxxx 11 3e 2e 1e 0f 07 03 01 000087f-00800 0000010000xxxxxxx 11 30 20 10 08 04 02 01 00007ff-007c0 00000011111xxxxxx 10 2f 1f 0f 07 03 01 00 00007bf-00780 00000011110xxxxxx 10 2e 1e 0f 07 03 01 00 000077f-00740 00000011101xxxxxx 10 2d 1d 0e 07 03 01 00 000073f-00700 00000011100xxxxxx 10 2c 1c 0e 07 03 01 00 00006ff-006c0 00000011011xxxxxx 10 2b 1b 0d 06 03 01 00 00006bf-00680 00000011010xxxxxx 10 2a 1a 0d 06 03 01 00 000067f-00640 00000011001xxxxxx 10 29 19 0c 06 03 01 00 000063f-00600 00000011000xxxxxx 10 28 18 0c 06 03 01 00 00005ff-005c0 00000010111xxxxxx 10 27 17 0b 05 02 01 00 00005bf-00580 00000010110xxxxxx 10 26 16 0b 05 02 01 00 000057f-00540 00000010101xxxxxx 10 25 15 0a 05 02 01 00 000053f-00500 00000010100xxxxxx 10 24 14 0a 05 02 01 00 00004ff-004c0 00000010011xxxxxx 10 23 13 09 04 02 01 00 00004bf-00480 00000010010xxxxxx 10 22 12 09 04 02 01 00 000047f-00440 00000010001xxxxxx 10 21 11 08 04 02 01 00 000043f-00400 00000010000xxxxxx 10 20 10 08 04 02 01 00 00003ff-003e0 000000011111xxxxx 9 1f 0f 07 03 01 00 00 000021f-00200 000000010000xxxxx 9 10 08 04 02 01 00 00 00001ff-001e0 000000001111xxxxx 8 0f 07 03 01 00 00 00 000011f-00100 000000001000xxxxx 8 08 04 02 01 00 00 00 00000ff-000e0 000000000111xxxxx <8 07 03 01 00 00 00 00 000001f-00000 000000000000xxxxx <8 00 00 00 00 00 00 00 001000-107ff 100000xxxxxxxxxxx 15 f0 e0 d0 c0 b0 a0 90 8010800-10fff 100001xxxxxxxxxxx 15 f1 e1 d1 c1 b1 a1 91 8011000-117ff 100010xxxxxxxxxxx 15 f2 e2 d2 c2 b2 a2 92 8117800-17fff 101111xxxxxxxxxxx 15 ff ef df cf bf af 9f 8718000-183ff 1100000xxxxxxxxxx 14 e0 d0 c0 b0 a0 90 80 8818400-187ff 1100001xxxxxxxxxx 14 e1 d1 c1 b1 a1 91 80 8818800-18bff 1100010xxxxxxxxxx 14 e2 d2 c2 b2 a2 92 81 881bc00-1bfff 1101111xxxxxxxxxx 14 ef df cf bf af 9f 87 8b1c000-1c1ff 11100000xxxxxxxxx 13 d0 c0 b0 a0 90 80 88 8c1c200-1c3ff 11100001xxxxxxxxx 13 d1 c1 b1 a1 91 80 88 8c1deff-1dfff 11101111xxxxxxxxx 13 df cf bf af 9f 87 8b 8d1e000-1e0ff 111100000xxxxxxxx 12 c0 b0 a0 90 80 88 8c 8e1e100-1e1ff 111100001xxxxxxxx 12 c1 b1 a1 91 80 88 8c 8e1ef00-1efff 111101111xxxxxxxx 12 cf bf af 9f 87 8b 8d 8e1f000-1f07f 1111100000xxxxxxx 11 b0 a0 90 80 88 8c 8e 8f1f800-1f0ff 1111100001xxxxxxx 11 b1 a1 91 80 88 8c 8e 8f1f780-1f7ff 1111101111xxxxxxx 11 bf af 9f 87 8b 8d 8e 8f1f800-1f83f 11111100000xxxxxx 10 a0 90 80 88 8c 8e 8f 8f1f840-1f87f 11111100001xxxxxx 10 a1 91 80 88 8c 8e 8f 8f1f880-1f8bf 11111100010xxxxxx 10 a2 92 81 88 8c 8e 8f 8f1f8c0-1f8ff 11111100011xxxxxx 10 a3 93 81 88 8c 8e 8f 8f1f900-1f93f 11111100100xxxxxx 10 a4 94 82 89 8c 8e 8f 8f1f940-1f97f 11111100101xxxxxx 10 a5 95 82 89 8c 8e 8f 8f1f980-1f9bf 11111100110xxxxxx 10 a6 96 83 89 8c 8e 8f 8f1f9c0-1f9ff 11111100111xxxxxx 10 a7 97 83 89 8c 8e 8f 8f1fa00-1fa3f 11111101000xxxxxx 10 a8 98 84 8a 8d 8e 8f 8f1fa40-1fa7f 11111101001xxxxxx 10 a9 99 84 8a 8d 8e 8f 8f1fa80-1fabf 11111101010xxxxxx 10 aa 9a 85 8a 8d 8e 8f 8f1fac0-1faff 11111101011xxxxxx 10 ab 9b 85 8a 8d 8e 8f 8f1fb00-1fb3f 11111101100xxxxxx 10 ac 9c 86 8b 8d 8e 8f 8f1fb40-1fb7f 11111101101xxxxxx 10 ad 9d 86 8b 8d 8e 8f 8f1fb80-1fbbf 11111101110xxxxxx 10 ae 9e 87 8b 8d 8e 8f 8f1fbc0-1fbff 11111101111xxxxxx 10 af 9f 87 8b 8d 8e 8f 8f1fc00-1fc1f 111111100000xxxxx 9 90 80 88 8c 8e 8f 8f 8f1fde0-1fdff 111111101111xxxxx 9 9f 87 8b 8d 8e 8f 8f 8f1fe00-1fe1f 111111110000xxxxx 8 80 88 8c 8e 8f 8f 8f 8f1fee0-1feff 111111110111xxxxx 8 87 8b 8d 8e 8f 8f 8f 8f1ff00-1ff1f 111111111000xxxxx <8 88 8c 8e 8f 8f 8f 8f 8f1ffe0-1ffff 111111111111xxxxx <8 8f 8f 8f 8f 8f 8f 8f 8f__________________________________________________________________________

The foregoing description of the preferred embodiment of the invention has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed, and many modifications and variations are possible in light of the above teaching. The preferred embodiment was chosen and described in order to best explain the principles of the invention and its practical applications to thereby enable others skilled in the art to best utilize the invention and various embodiments and with various modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined only by the claims appended hereto.

Claims

1. A dynamic digital audio filter comprising

adding means for receiving an input audio signal,

first and second unit delay means connected to an output of said adding means for forming first and second unit output delays each having a first bit width, respectively, wherein the first unit delay is one time unit and wherein the second unit delay is two time units,

means for providing a first unsigned coefficient and a second unsigned coefficient,

unsigned multiplying means receiving said first and second output delays, wherein the inputs to said multiplying means are a truncation of said first and second output delays each having a truncated bit width such that the truncated bit width is less than said first bit width so as to form a first truncated delay and a second truncated delay,

said unsigned multiplying means including means for providing said first unsigned coefficient times the first truncated delay and second unsigned coefficient times the second truncated delay,

means for shifting the output of said first delay means to multiply by two,

means for forming a 2's complement of the output of said second delay means, and

said adding means including means for forming an output signal which is a sum of said input audio signal, the output of said first delay means shifted to multiply by two, the 2's complement output of said second delay means, the first unsigned coefficient times the first truncated delay, and the second unsigned coefficient times the second truncated delay.

2. A digital filter as in claim 1 including means for forming a difference between said first delay and said input audio signal and wherein said sum includes the first unsigned coefficient times the difference between said first delay and said input audio signal and including means for forming a difference between said second delay and said input audio signal and wherein said sum includes the second unsigned coefficient times the difference between said second delay and said input audio signal.

3. A digital filter as in claim 1 further including a finite impulse response filter.

4. A digital filter as in claim 1 including first storage means and wherein said second coefficient is stored in a logarithmically compressed format in said first storage means.

5. A digital filter as in claim 4 including second storage means and including a third coefficient stored in said second storage means in a logarithmically compressed format representative of a difference between said first and second coefficients.

6. A digital filter as in claim 5 including means for interpolating between said first and second coefficients.

7. In a dynamic digital audio filter for filtering an input audio signal wherein said filter includes first and second unit delays for forming first and second unit output delays having a first bit width wherein the first unit delay is one time unit and wherein the second unit delay is two time units, an unsigned multiplying means wherein an input to said multiplying means is a truncation of said first and second output delays each having a truncated bit width such that the truncated bit widths are less than said first bit width so as to form first and second truncated delays, an adding means, the method comprising the steps of

providing a first unsigned coefficient and a second unsigned coefficient, providing said first unsigned coefficient times the first truncated delay and said second unsigned coefficient times the second truncated delay,

receiving said first and second truncated delays,

shifting the output of said first unit delay to multiply by two,

forming a 2's complement of the output of said second unit delay, and

forming an output delayed signal which is a sum of said input audio signal, the output of said first unit delay shifted to multiply by two, the 2's complement output of said second unit delay, a first unsigned coefficient times the first truncated delay, and a second unsigned coefficient times the second truncated delay.