All-pass filter problems


2004-08-13 1:01 pm
Hi all

I’m using 16 stages of allpass filters to create a 0.8ms delay in order to compensate de distance between a bass speaker and a horn mounted driver. A LM1876 is used for amplification.

Allpass filters are all the same, R1 then C1 on the + input, R2 on the – input, R3//C3 on the feedback. R1=23.7k, C1=1nF, R2=R3=22k, C3=22pF. OpAmps are LM837
At the end, the signal has a lot of distortion, I can see it with the scope !!
I tried with TL074 opamps and suppressed the 22pF caps.
Distortion is lower but not enough, noise is very higher than with LM837………
I saw at Linkwitz website that he uses an R and C filter, the next one is a C and R filter and so on…
Is there a reason for that ?
What could I do (staying with analog delays, no BBD, no digital) to get low distortion and low noise, without risk of instability of the opamps ?

Thanks for your help
Best Regards


Paid Member
2007-07-17 2:35 am
Central Berlin, Germany

Up to which frequency do you need to delay? I assume you delay the bass channel (treble channel would be way more effort to delay). Of what type (frequency, order, characterstic) is the crossover?

Normally, an allpass which is used for delay should not be a simple cascade of 1st-order filters, you loose flat delay bandwidth with that and also you need more active stages than necessary. For example, to delay 2ms up to 1kHz only a 7th-order filter is needed.

For a delay, you want to have a Butterworth characteristic of the group delay, that is you want to keep the delay vs. frequency as flat as possible up the corner frequency and then a quick roll-off starts. With the simple cascade the group delay isn't as flat as it could be, but the roll-off is more gentle, which might be an advantage depending on the application.

This leads to a complex filter function, which can be realised with 2nd-order filter blocks, using two opamps per stage. If the overall order is odd, one 1st order stage is added.

See Texas Intruments AppNote Active Filter Design Techniques for details on allpass design.

And use pretty decent opamps of current production, LM4562 duals or LME49740 quads for example.

To find the correct delay one would certainly need to make measurements because the acoustic filter functions need to be taken into account. For evaluation a digital delay is a good thing to have, a PC with a wave editor will normally suffice for that (assuming you drive both horn and subwoofer actively and the crossover can be spilt into seperately driven low and high sections).

- Klaus


2004-08-13 1:01 pm
Hi Klaus

Thanks for your answer

I give you the details of the application (4x2ways amps for home cinema):
I delay the bass channel (Beyma 15MI100 sealed box) The High channel is Altec 806-8A
Active low pass filter : Bessel 3rd order 532Hz
Active high pass filter : Butterworth 2nd order 980Hz added phase 180°
Bass driver offset due to crossover : +70mm
(simulation done with “Filtre_simul” of J.M Lecleac’h )
Distance between cones = 345mm so I need 275mm that is 0.808ms delay

I used WinIsdProAlpha to simulate the bass driver (watching group delay) and added 16 All-pass filter of 0.05ms with the filter editor, I also draw the curve (excel) of Tg=(2xRxC)/(1+(f/f0)^2) to determinate the minimum time I could use on each filter and keeping a constant delay far from the cut-off frequency . One decade seems ok, but really I don’t know, maybe we don’t need so much ?
However, simulation in WinIsd is ok with 16x0.05ms : the group delay curve climbs “constantly” between 0 to 2kHz and it does not if I put values like 0.25ms

Now, it’s difficult (impossible, maths are as hell for me) , for me to draw and compare the summed responses of the [lowpass Bess3+ highpassButt2 + 16x0.05ms allpass 1st order] and a [lowpass Bess3+ highpassButt2 + 4x0.2ms allpass2nd order] for example.

I watched to 2nd order filters, but I don’t have precision smd caps and res to do it ……… I choose 1st order…….
Now, reading the “Texas instrument active filter design”, I don’t understand what they call the “normalised” group delay ?

My problem is that I soon did the 4 PCBs of the amplifiers with filters and delays.
I’m ok to use LME49740, but before buying them ($!!!) I’d like to be sure I really need them.
I usually use LM837 because them are cheep and not noisy (3.5nV/sqr(Hz)), but they need feedback caps if gain is low, and with these caps, after the 16th allpass, sine wave looks like sawtooth !

To be followed

Best regards
I recommend that you go to and download their free FilterPro software. It will design many types of filters, and draw and label the schematic for you, and plot the frquency reponse of the amplitude, phase, and group delay. When you need better simulations and plots, use LTspice, from .

Note, too, that for certain types of filters, TI's FilterPro will tell you the minimum speed required of the opamps, which can be critical to the proper operation of the filter.

A very good reference for active filter design (all types) is Walt Jung's "Op Amp Applications Handbook", available as a free download, in whole or in part, at .


Paid Member
2007-07-17 2:35 am
Central Berlin, Germany
Hi Bruno,

I downloaded the .xls of Mr. Lecleac'h, typing in your values and getting reasonable results. I found, by comparing the results with a LTspice simulation, that the 70mm offset can be ommitted, also giving only +-0.15dB ripple on axis, when we increase the LP cutoff to ~618Hz, and the spreadsheet confirmed this. The difference is in the "coincidence" curve which has +1dB ripple, can you elaborate what this curve exactly shows? To me, it looks like being the magnitude-only sum of the transfer functions, neglecting phase. Which is important for systems with a bigger distance between ways...

With 3rd order Bessel we are -60dB down with the woofer one decade up the lowpass roll-off, I would estimate this is more than sufficient. The 15MI100 looks pretty nice in the HF anyway, so I'd say some 40dB distance would be OK, giving us about x6 the corner frequency. With 600Hz we get 3.6kHz, delaying that 0.8ms gives us a normalized Tgr of 2.88 (0.8ms*3.6kHz), which, using the tables at the end of the TI document, tells us we need a 9th or 10th order allpass filter. I could simulate that for you with LTspice, which can directly operate on the laplace transfer equations (and work with time delays as well).

BTW, from a viewpoint of room exitation (polar response) I would always try to not introduce any additional factors, that is, get the cones acoustically coincidend for all radiation angles just by the way of mounting them. With a delay you can correct only for smooth on-axis response. Just my personal take, I don't want to critizise your design in any way.



2004-08-13 1:01 pm
Hi Tom & Klaus

What is also very interesting in .xls of Mr Lecleac’h (last version) is the 100Hz square wave curve. It gives an “idea” of phase/impulse response.
You can find this .xls file “” with 100Hz square wave response at
at about 60% down in this page.

When using this .xls , I tried to keep a high value of bass driver offset in order to minimise the needed delay to make coincidence of the speakers cones. While doing this, the 100Hz square wave curve must be “nice” (good rising edge, flat, no overshoot …)
85 mm and 540Hz could be also good values.
I tried many horns I have, to get good polar response of the couple 15MI100/806-8A and chose (only with my ears…) rcf H4823 model.

The “middle” frequency of the 2 crossovers is about 727 Hz, so as you say, with 40dB distance, Tgr=3.5ms
I will try 12th order filter instead of 16 what is easy to do by keeping the same PCB (only 3xQuad opamp) but I can’t built 2nd order filter with this PCB (0806 smd are too small !)
So, I’m looking for others opamps : 4 in the same 14dil package, low noise, stable at unity gain, suitable for this application, correct price. If I don’t find, I’ll try 2xLT1229 soic (I have many of them) with adaptation board to convert to 14dil.
Because of the big number of opamp, I decided to inject a 8Vrms signal at the input to increase s/n ratio. Using “Visual Analyser 8” , measured distortion was about 0.6% with only odd harmonics (H3, H5, H7) . Now, injecting a 0.5Vrms signal, distortion is only 0.1%.
Does it come from a slew-rate problem?

I downloaded the FilterPro software at; yes very usefull but no allpass filter :bawling:
It is quite the same than the on line one : (needs java)

I don’t have much time to check longer these days, I’ll do during week-end.

Tks & Rgds