Добавил:
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
акустика / self_douglas_the_design_of_active_crossovers.pdf
Скачиваний:
226
Добавлен:
04.05.2023
Размер:
60.48 Mб
Скачать

Time-Delay Filters  373

Figure 13.26: Distortion plot for the 2nd-order stage output (A2) of the 3rd-order 80 usec allpass filter in Figure 13.22, when it is loaded by the RC or CR final stage, and also with no load (NL). Polypropylene capacitors, 5532s, input 9 Vrms.

is now peaking in delay, the MFB filter has a higher Q and therefore a higher passband gain, here 0 dB. Thus R3 is half the value of R4, R5 instead of a quarter. Therefore the noise gain in the subtractor stageA2 is 3.6 dB lower at 12.0 dB.

This is another reason why a 3rd-order allpass is better than a 2nd-order allpass delay filter; it is not only flatter, it is quieter. Is this an original observation? I suspect so.

Higher-Order Allpass Filters

If 3rd-order filters do not provide a sufficiently extended flat-delay characteristic, then higher-order filters might be considered. We have just seen that a 3rd-order allpass filter provides the most extended flat-delay response for a given component count and also has usefully lower noise. If greater delays or flatter delay characteristics than this configuration can provide are required, higher-order filters may be the answer.

The basic principle, as we saw for the 3rd-order filter, is to take one or more 2nd-order stages with a high Q, giving a peak in the delay response, and then cancel out the peak by cascading it with one or

374  Time-Delay Filters

Figure 13.27: Distortion plot for the 2nd-order-stage output of the 3rd-order 80 usec allpass filter in Figure 13.22 loaded only by the CR final stage. Adding the 3k3 output biasing resistor reduces distortion below 6 kHz. Polypropylene capacitors, 5532s, input 9 Vrms.

more stages, either 1stor 2nd-order, with a low Q. Figure 13.28 shows how higher-order filters are built up. For each stage the Q is that of the overall allpass function and not the Q of the MFB filter, which is a different quantity. For example, in a 2nd-order allpass filter the Q for maximal flatness is 0.58, but the Q of the MFB filter is 0.50. Here the 1st-order sections have been put at the end of the chain; this makes no difference, as all the stages have unity gain.

Designing high-order filters is not something to try by hand, and the usual design procedure is to consult a table of allpass filter coefficients, as in Table 13.4. This assumes you are using the 1–2BP approach to making allpass filters.

To design your filter, pick an order and the cutoff frequency f required. For each stage of the complete filter, first select the value of C = C1 = C2 in the MFB filter (here I am using the component designations in the first stage of Figure 13.20); if this results in resistor values that are undesirably low or high, then you will have to go back and rethink the value of C. R1 and R2 are then calculated using the coefficients ai and bi thus:

R1 =

ai

13.8

4π fC

 

 

 

Time-Delay Filters 

375

 

 

 

 

 

R2 =

 

bi

13.9

a

π fC

 

 

i

 

 

 

This sets the frequency and Q of the stage. We now need to work out the parameter α:

α =

(ai )2

13.10

b

 

 

 

i

 

which lets us set R3 at R/α, where R is the value of R4 and R5, to get the correct subtraction to create an allpass filter; parameter α includes the 2 in the 1–2BP. Job done, but your very next move should be to run the complete filter on a simulator to make sure you get what you want—it’s all too easy to make an arithmetical error.

Figure 13.29 shows the delay response of 1stto 6th-order allpass filters all set to the same cutoff frequency. This is the sort of diagram you often see in filter textbooks, but here it is simulated rather than derived mathematically.

Figure 13.28: Constructing allpass filters from 3rdto 6th-order by cascading 1stand 2nd-order stages. The relative cutoff frequency and the Q are shown for each stage.

Table 13.4: Filter coefficients for allpass filters (after Kugelstadt).

Filter

Filter

Coeff

Coeff

Section Q

R3 factor

Stage gain

 

 

 

 

 

 

 

Order

Section i

ai

bi

Q

 

 

1st

1

0.6436

n/a

 

 

1.000

2nd

1

1.6278

0.8832

0.58

3.000

0.667

3rd

1

1.1415

n/a

 

 

1.000

 

2

1.5092

1.0877

0.69

2.094

0.955

4th

1

2.3370

1.4878

0.52

3.671

0.545

 

2

1.3506

1.1837

0.81

1.541

1.298

5th

1

1.2974

n/a

 

 

1.000

 

2

2.2224

1.5685

0.56

3.149

0.635

 

3

1.2116

1.2330

0.92

1.191

1.680

6th

1

2.6117

1.7763

0.51

3.840

0.521

 

2

2.0706

1.6015

0.61

2.677

0.747

 

3

1.0967

1.2596

1.02

0.955

2.095

 

 

 

 

 

 

 

Figure 13.29: Delay response of 1stto 6th-order allpass filters, all set to same cutoff frequency. The delay increases with filter order.

Time-Delay Filters  377

More relevant to our needs is to set the group delay to the same value (80 usec again) for each order of filter, and you can see how much more extended the flat portion of the delay curve is with the higherorder filters, as in Figure 13.30. It is also painfully obvious that the extra flat-delay bandwidth you gain with each increase of filter order decreases; this is true on a log-frequency plot because with each higher order you actually gain a constant extra bandwidth of about 5 kHz. There is clearly a limit to how high an order it is sensible to use.Apowerful consideration here is that component sensitivity (i.e. how much the filter behaviour changes with component value tolerances) gets worse for higher-order filters. There are also issues with dynamic range, as higher-order allpass filters use at least one section with high Q—this means its MFB filter has gain well above unity, and this is a potential clipping point. The 6th-order allpass filter described later illustrates this problem.

I will now give some practical designs for 4th-, 5th-, and 6th-order allpass filters, all set for a 80 usec LF delay; these were the circuits used to generate the curves labelled 4, 5, and 6 in Figure 13.30. The capacitor values have been chosen to keep the resistor values in the MFB filters above 1 kΩ to minimise loading problems. The resistor values given are exact and not part of any series of preferred values.

Figure 13.31 shows a 4th-order allpass filter made by cascading two 2nd-order allpass stages. The delay is now 10% down at 17.2 kHz, rather than the 12.7 kHz we got with the 3rd-order filter examined earlier in this chapter, a useful increase in flat-delay bandwidth.

The lower the Q of the stage, the lower the gain of the MFB filter, and so the lower the value of the resistor feeding into the summing stage must be to get the correct gain. It must be remembered that this resistor constitutes a load to ground on the output of the MFB opamp, and its value must not be

Figure 13.30: Delay response of first to 6th-order allpass filters, all set to 80 usec delay. The 72 usec line shows where each response is 10% down.

378  Time-Delay Filters

Figure 13.31: Fourth-order allpass filter built by cascading two 2nd-order stages. Delay = 80 usec.

Figure 13.32: Fifth-order allpass filter built by cascading two 2nd-order stages and one 1st-order stage. Delay = 80 usec.

allowed to fall below 500 Ω if 5532 or similar opamp types are being used. In the first stage here this means that R4 and R5 have to be 2 kΩ to get R4 up to 544.8 Ω. In the second stage the Q is higher and the MFB has more gain, so R9, R10 can be reduced to 1 kΩ for better noise performance, while R8 is a respectably high 648.9 Ω.

The values of C1, C2 and C3, C4 can be selected independently for each 2nd-order stage, and C3, C4 have therefore been set lower in value, at 4n7, to keep R6, R7 above 1 kΩ in value.

Figure 13.32 shows a 5th-order allpass filter made by cascading two 2nd-order allpass stages and one 1st-order stage. Because of the higher cutoff frequency, both C1, C2 and C3, C4 have been set to 4n7 to keep the associated resistors above 1 kΩ. C5 can be set independently and has been made 10 nF to give a suitable value for R11: as low as possible for noise purposes, but not so low that it puts an excessive load on A4.

Figure 13.33 shows a 6th-order allpass filter made by cascading three 2nd-order allpass stages. In the third high-Q stage it has been necessary to reduce C5, C6 to 3n3 to keep the associated resistors above 1 kΩ.

The final 2nd-order section has a Q of 1.02, which means a passband gain of 2.095 times in the MFB filter. This could cause headroom problems. It is another reason not to embrace higher-order allpass filters without careful consideration.