Fullscreen HTML5Med Res (??MB)HTML4

AUDIO OUT AUDIO OUT L R By Jake Rothman All-pass filters, part 5 – phase-linear crossover and more Over the last few issues, we’ve been looking into all-pass filters, ie, those that affect the phase response but not the frequency response of a signal. We’ve even presented multiple circuit boards to implement practical all-pass filters for practical purposes, like loudspeaker crossover time alignment and making audio signals more symmetrical so they can be recorded or transmitted at a higher level (louder). Phase-linear crossover board As promised, I have now had a board made for the subtractive phase-linear crossover circuit (Fig.33) described in part three of the all-pass series, in the June 2026 issue. I’ve added a few stability components (resistors R17 & R18, plus capacitors C9, C10 & C12) as well as decoupling capacitors R1 2.2kΩ +20 6 C5 100nF C4 100nF – 1 5 C16 100nF C3 100nF Linear-phase crossover frequency response +25 0V 8 IC1a 3 5532 + 4 R2 2.2kΩ +30 C14 100µF 25V R3 2.2kΩ R4 2.2kΩ Low-pass output – IC1b 5532 + Relative Amplitude (dBr) Input + 2 C2 100nF a phase-linear crossover in 1987, which was a three-way design fixed at 500Hz and 5kHz. It had a sizable impact at the time. The idea was that the relative phase between the outputs did not shift with frequency. The result is supposed to be a small, constant delay across the whole audio band. In the Elektor tradition of “form before function”, the design had a difficult PCB layout with thin tracks that fell off during soldering. It also had an onboard power supply using an unobtainable LM325 dual-voltage regulator. The critical capacitors were weird Dutch Philips polystyrene types with diagonally placed pins. I later discovered that LCR in Wales makes a compatible type. Despite the difficulties, of course, I had to build one (shown in Photo V+ Fig.52: the phaselinear crossover filter circuit. C1 100nF C14, C15, C16, C17 & C18 to the original design. There was one spare op amp (IC3b) in a dual package left over, so this was used for a polarity inversion circuit. The new, complete circuit is given in Fig.52. The curve is that of a classic fourth­order Linkwitz-Riley (LR) crossover with a -6dB dip at the crossover point of 500Hz, as shown in Fig.53. Incidentally, I find it odd that someone can have a circuit that is simply two cascaded Butterworth filters named after them. I guess it’s the analysis that followed that was so significant. [Editor’s note: consider that the response of two cascaded Butterworth filters is not the same as a second-order Butterworth filter response.] Elektor magazine published the first home constructor’s design for 7 R18 47Ω C6 100nF +15 +10 +5 +0 -5 -10 -15 -20 + V– -25 C15 100µF 25V -30 0V R11 1kΩ C9 47pF R5 2.2kΩ C8 100nF R7 4.3kΩ 2 6 C12 47pF R16 1.2kΩ R10 1kΩ – + 7 2 8 IC3a 3 5532 + 4 – C17 100nF 0V V– 6 + C13 100µF R17 47Ω – IC3b 5 5532 R13 100kΩ V– R15 1.2kΩ 1 C18 100nF R6 2kΩ 66 C11 100µF + IC2b 5 5532 5k 10k 20k R12 1kΩ R9 2kΩ R8 1kΩ 1 C10 47pF + 8 IC2a 3 5532 + 4 – 50 100 200 500 1k 2k Frequency (Hz) Fig.53: the phase-linear crossover frequency response, a standard fourth-order Linkwitz-Riley at 500Hz. V+ V+ C7 100nF 20 7 R14 1.2kΩ High-pass output S1 High-pass polarity switch – closed for invert 0V Practical Electronics | August | 2026 Delay Ouputs LP Input Bass Two-way crossover 1 HP Extra delay Fig.54: making a three-way crossover out of two two-way boards. Watch out for the extra delay introduced by the second board relative to the bass output. LP Mid HP Tweeter Two-way crossover 2 Low-pass LR 4th-order All-pass 2nd-order inverting fc = 300Hz fc = 3kHz Low-pass output Input Scope 1: the low-pass & high-pass signals remain in phase during a sweep through the crossover region. R R All-pass 2nd-order inverting – Low-pass LR 4th-order Band-pass (mid) output + R fc = 300Hz Subtractor fc = 3kHz R Fig.55: the original Elektor three-way configuration gave a constant delay for all outputs. 0V R R All-pass 2nd-order inverting – High-pass output R fc = 3kHz + 0V 42) and found it worked well with a Philips 50mm hard-paper dome midrange driver, a 10-inch (250mm) bass unit and a one-inch (25mm) Mylar dome tweeter. The use of high-value resistors, LF353 op amps, and the delay and subtraction stages all made it very hissy, especially the tweeter output. I tried running it at +6dBu while attenuating the output to the power amps, but it was still bad. I ended up binning it after the umpteenth track breakage. Obviously, a board redesign was called for. I thought a two-way board would be the simplest to start with. I did think another could be cascaded if a three-way design was needed, as shown in Fig.54, but the mid/tweeter would then be delayed relative to the bass. However, with a physically displaced woofer, this could be a good thing. For a proper three-way crossover with a consistent delay, a new three-way board will have to be designed using the block diagram shown in Fig.55, along the lines of the original Elektor circuit. However, I won’t bother doing that now, for reasons given below. Relative phase In theory, the high-pass and lowpass outputs should remain relatively in-phase as the frequency is Practical Electronics | August | 2026 Summer swept through the crossover point. Using a dual-trace oscilloscope showed this to be the case – see Scope 1. This is not like using separate high-pass and low-pass filters, where there is a phase difference between the two outputs. I then tested a state-variable filter, which also employs subtraction. Interestingly, it had the same lack of relative phase shift between the high-pass and low- Scope 2: the effect of a high-pass filter on a 500Hz square wave. The state variable gave the same result. pass outputs. This makes it a viable competitor to the phase-linear filter. I thought maybe the group delay would be more variable with frequency; the phase-linear filter is meant to be constant. I looked at the output waveforms with square waves and found both filters to give the same resulting waveshapes, shown in Scope 2. This suggests that the phase response and therefore the group Photo 42: the Elektor phaselinear crossover board. Its tracks were too thin for a single-sided board. 67 Output amplitude (dB) Tweaks Going back to separate high-pass and low-pass filters based on the Sallen and Key configuration would need eight capacitors. However, this offers a great advantage which the state-variable and phase-linear circuits lack: control of the overlap at the crossover point. Table 1 – frequency-determining crossover capacitor selection +30 Effects of altering the filter subtraction level +25 Frequency Capacitor value 5kHz 10nF 4kHz 12nF 3kHz 15nF 2kHz 27nF 1kHz 47nF 700Hz 68nF 600Hz 75nF 500Hz 100nF -25 300Hz 150nF -30 250Hz 220nF 125Hz 390nF 100Hz 470nF +20 Relative Amplitude (dBr) delay were the same for both filters. For a fourth-order filter, the state-variable uses four frequency-­ determining capacitors, whereas the phase-linear uses six, which makes it slightly more expensive to do the same job. It is also more susceptible to component tolerances. As is sometimes the case, when I try a new wonderful circuit trick based on papers in the Journal of the Audio Engineering Society, full of complex (in both senses of the word) mathematics, I find there is little to be gained compared to standard textbook stuff being knocked out by Chinese manufacturers, such as the £150 Behringer state-variable based crossover filter I compared it to. A photo of that unit was shown in Photo 12 in part two of this series, in the May 2026 issue, on page 55. Marketing and technical hype must always be personally tested. With a true phase-linear filter, the outputs would sum to the original input waveform. For both filters, the output is an all-pass function. Still, I had 30 boards made, so they will have to make themselves useful; they just need two more capacitors. They will also provide an experimental platform for investigating subtraction techniques, and the second-order all-pass network could be useful for something like an EQ dip. +15 +10 +5 +0 -5 -10 -15 -20 20 50 100 200 500 1k 2k Frequency (Hz) 5k 10k 20k Fig.56: the effect of altering the subtraction level in the filter. Yes, there are phase problems, but the final acoustic result is superior to the hole-in-the-middle effect invariably produced by the classic Linkwitz-Riley state-variable filter (a result of its -6dB dip) using real drivers with natural roll-offs. [Editor’s note: In theory, a Linkwitz-Riley crossover gives ideal on-axis summing, although the off-axis response may not be perfect. Still, real drivers rarely behave like textbook sources. Other alignments may sacrifice some on-axis perfection while improving the perceived integration or off-axis balance. It will depend on the drivers, layout and listener preference.] Frequency scaling For the 500Hz crossover frequency, all the frequency-­d etermining capacitors are 100nF. Some of the filter capacitors are doubled up, such as the feedback capacitors C1/C2 and C4/C5. This is because sometimes the doubled values, such as 200nF, are from the E24 series and so harder to obtain. There’s also the advantage that if all the 100nF capacitors are from the same batch, they’re likely to be much closer in value to each other than their tolerances might suggest. The equal-value Sallen and Key circuit is not used because it has gain in the passband, increasing the noise level. The resistors are all kept the same and as low in value as possible for minimum Johnson (thermal) noise. To change the frequency, scale the capacitors accordingly – see Table 1. These are all rounded off to the nearest E12 value, so the frequencies are somewhat approximate. Nice curves R9, the subtraction mixing resistor, is the most critical component. Fig.56 shows the effect of using 1.8kΩ and 2.4kΩ values for R9. If not enough signal is subtracted (1.8kΩ), the high-pass action just stops and it flattens off. Too much (2.4kΩ) and a notch develops, with the signal coming back with a polarity reversal. Tweaking R9 can produce a useful curve akin to that used in the KEF 104aB speaker for the 10 0 –10 –20 –30 –40 20 50 100 200 500 1k Frequency (Hz) 2k 5k 10k 20k Fig.57: KEF’s 104aB tweeter high-pass filter. Notice the anti-resonance notch and the reversion of the third-order slope to first-order afterwards. 68 Practical Electronics | August | 2026 High-pass High-pass V+ 0V 0V V– Low-pass Low-pass C2 R 1 C3 R 4 J1 50 100 200 500 1k 2k Frequency (Hz) tweeter described in Audio Out, June 2026. The response of this third-order high-pass filter is shown in Fig.57. This takes into account the second-order highpass slope that the T27 tweeter exhibits after resonance, making the total electroacoustic output a smooth third-order. The dip to suppress the tweeter’s fundamental resonance at 1.2kHz can be seen, after which the curve slope reduces to -6dB/ octave. These types of filters work well with satellite speakers with a pronounced bass resonant hump, such as the LS3/5a, in subwoofer applications. Also, dome-type mid-range drivers often have high-Q resonances around 350Hz, such as the ATC 3-inch dome and some Philips units. Some dome tweeters with skimpy neodymium magnets, such as the DT28, often roll off with a peak at 2kHz. They will also benefit from a bit of a notch. Remember that the reduced stop-band attenuation will reduce power handling and increase distortion. Distortion measurements I have tried measuring the distortion of filters with the THD+N (total harmonic distortion plus noise) measurement macro supplied with the Audio Precision system, but it gave rather strange results. The main problem is that as the filtering action takes place, there is less and less signal for the distortion analyser to work on. This gives the false impression that the distortion is rising excessively rapidly as the analyser ramps up the gain to compensate. The effective result is a boosted noise component of the residual signal, giving falsely high reading. This test is designed for amplifiers, which have a flat response. It helps if the filter is driven at a Practical Electronics | August | 2026 R 18 High-pass output C17 R 8 C8 5k 10k 20k Fig.58: the artificially high distortion for the crossover filter at 8V RMS. C6 R2 R 6 20 C5 C9 Input .002 .001 .0005 C4 IC1 Input 0V 0V Output J2 R3 C16 J5 C1 0.02 0.01 .005 Low-pass output C14 + 0.2 0.1 0.05 .0002 .0001 C15 Power + 2 1 0.5 R 5 IC2 C18 C10 C7 R 7 R 11 IC3 R 9 R 10 C12 R 12 0V 0V Output J3 R 15 R 16 C13 R 17 Closed for invert S1 J4 + Total Harmonic Distortion (%) Linear-phase crossover filter THD+N vs freq. + 10 5 C11 R 13 R 14 High-pass polarity switch Fig.59: the PCB overlay for the phase-linear crossover. Note the multiple outlines available for the frequency-determining capacitors. high level and a bit of inverse pre-­ Assembly filtering is applied to flatten it out. This is a conventional, mature Of course, all filters do have 1980s audio technology design a real rise in distortion in the with no fiddly surface-mount turnover region, where the main resistors, just proper blue thinfilter slope takes place between film axial types. There is plenty the passband and the stopband. of space for different shapes and This is because there is voltage sizes of frequency-determining developed across the capacitors, capacitors; six mounting holes a higher common-mode voltage are provided for each. The PCB on the op amp input pins, and overlay is shown in Fig.59 and the positive feedback to compensate completed assembly in Photo 43. for resistive losses. One of my friends had bags of The question is how to measure 16nF Suflex polystyrene capacithe filter THD accurately, which I tors, which are very large for their can’t do at the moment. The result value, and they fitted fine. is shown in Fig.58, using NE5532 op amps and polystyrene capacitors. Using the board When I look at the residual on the It should go without saying that scope or do an averaged FFT, the the filter should be driven from a distortion is much lower than indi- low impedance source, since its cated in the graph. The distortion in the passband is very low; almost that of the op amps alone. Taking the same measurements at 1V RMS gave higher distortion readings, showing it is a noise effect. Photo 43: the completed PCB. Note the large, clear epoxy potted Rifa polystyrene caps. 69 Photo 44: boxing up the all-pass in an old satellite receiver 1U case. There was a 15V-0-15V toroidal transformer already installed. The phase-linear crossover can be treated similarly. input impedance is around 800Ω. This driver can be a voltage follower or a differential amplifier for input balancing. For minimum noise, drive the board as hard as possible, making sure the power amps it feeds clip first. The output of the board will often need considerable attenuation to achieve this, since most power amps are fully driven at 775mV RMS or thereabouts. Active speaker systems often suffer from multiple Earth loops. Earth isolation can be obtained by using balancing transformers at the outputs. I used 3:1 step-down transformers to get lower current drive requirements and attenuation at the same time. When selecting the crossover frequency, the drivers should be flat to about an octave away from the crossover frequency to avoid frequency response dips. The only way to get crossovers right is to actually measure the combined acoustic output, which is a lot of work. Boxing clever I will now get back to the project we described last month, the adjustable all-pass filter using an eight-gang pot. We finished the article last month without having put it in a case. This advice will also apply to the phase-linear crossover, since its case requirements are much the same. Fox boxing up the all-pass system, I used a 1U case from an old satellite receiver, as shown in Photo 44. There was a 15-015V toroidal transformer already installed and an edgewise signal level meter (sadly broken). Its slot will have to be filled by an illuminated logo in Quad style. This is another bit of audio electronics that drives me mad. Like all studio engineers, I prefer my gear in 19-inch (483mm) rack cases, but they are becoming expensive and difficult to get. For some reason, 1U high (44.45mm, 1¾-inch) cases are easier to find, but they have insufficient internal headroom for things like toroidal transformers, heatsinks and old-format XLR sockets. As shown in Photo 45, the XLRs have to go on their side because there is usually a lip from the lid and base. One way around it is to use Neutrik PCB-mount connectors. Q-max punches (Photo 46) are excellent for enlarging existing holes. The old format XLR sizes are 25mm for female sockets and 19mm for the male types. New format panel-mount sockets (Photo 47) are all 23mm. Always drill the 3mm mounting holes after the socket has been placed in the hole so that they line up. 2U cases are vastly superior for audio circuits, being 2.5 inches (63.5mm) tall, giving good front panel space for proper ‘human-sized’ knobs. You will have to give £87 to RS for a good aluminium one though. I balk at the cost; it’s often more than the electronics inside! I’m going to join forces with Grindle and other engineers to get a load made based on extruded Photo 45: I was disappointed that traditional XLR sockets wouldn’t fit normally on this 1U case; they had to go sideways. Still, it only cost £3.00! Photo 46: “Q-max” punches give a much cleaner result than drilling. Old guys have big collections from the valve era, so they’re cheaper at rallies. Photo 47: the new XLRs introduced by Neutrik take up less room and require identical holes. The trouble is I still have hundreds of the old types! 70 Practical Electronics | August | 2026 R4 2.4kΩ C3 10nF R1 1.3kΩ Input Low Zin 1.3kΩ R2 1.3kΩ R3 1.3kΩ 2 3 C1 130nF Photo 48: the old JLM power supply board is a very convenient design. I’ll design my own soon. aluminium corner pieces. It’s always worthwhile scouring radio rallies for old equipment with nice metal cases. Often, only a new front panel is required and a bit of hole punching at the back. I got the case I used for this project at a radio rally for just £3! Aluminium boxes are best if you are having the mains transformer in the case, since steel can spread its magnetic field around, increasing the risk of hum. Conversely, steel cases can minimise hum by their magnetic screening effect if the transformer is outside. In studios, we always try to have equipment with internal power supplies, though, because multiple plugpacks make an awful mess with the cabling and degrade overall system reliability. For the power supply in Photo 48, I used an old kit board from the Australian company JLM Audio, with a standard LM317/LM337-based circuit. Not again! As usual, doing this one measurement has opened a whole new area of investigation. Apparently, the most popular filter configuration, the Sallen and Key used here, is one of the worst for distortion. Ton Giesberts, Elektor’s chief audio designer, found that the inverting multiple-feedback type or Rauch filter is almost 10× better (Summer Circuits, July-­August 2005). These filters are not used much because it’s hard to calculate the component values, although this process is described in the National Semiconductor Audio/Radio handbook (section 5, page 2). They were used in the ETI Active Speaker by Barry Hughes and Paul Hynes years back in September 1982, and a crossover based on this is shown in Fig.60. [Editor’s note: I prefer multiple-­ feedback filters over Sallen-Key as Practical Electronics | August | 2026 Fig.60: the interesting inverting third-order Rauch crossover has a -3dB dip at 2.2kHz. I plan to investigate this configuration further. C2 110nF 8 – +15V IC1a 5532 1 Output + 4 –15V 0V 0V –3dB <at> 2.2kHz, 3rd-order low-pass filter C7 13nF R7 1.2kΩ Input Low Zin 600Ω C4 13nF R5 1kΩ C5 13nF R6 1.2kΩ C6 13nF 6 5 – IC1b 5532 7 Output + 0V 0V –3dB <at> 2.2kHz, 3rd-order high-pass filter they provide more effective attenuation at HF. Coincidentally, this is explained in Circuit Surgery this month – see page 44]. I thought the ETI Active Speaker was one of the best magazine active speakers I’ve built and I learnt a lot. Op amps in inverting configurations can give lower distortion because there is no common-mode voltage signal on the two input pins. They do suffer higher noise and lower input impedance in this configuration, though. I’m going to use a Rauch configuration in an all-pass filter to form a subtractive crossover to see how that works out. Phased out? Yes there’s more all-pass filter stuff we could do, obviously including a guitar phaser pedal, but I need to give it a rest for a while. Later on, we’ll use all-pass filters in column speakers, multiple tweeter arrays and the D’Appolito dual-midrange system with the tweeter in the middle. The Quad ELS63 electrostatic speaker used concentric rings fed by a tapped series of all-pass filters. Some public address speaker systems even use phased-array steering, a technique borrowed from radar and antenna design. I’m looking to use modulated all-pass filters to make a ‘rotating’ Leslie loudspeaker! There’s scope for employing all-pass filter delay lines for ‘look-ahead’ control in limiters, de-essers and pop filters, where the processing comes in before the audio peak arrives. The BBC created some interesting designs in the 1960s using this technique. Possibly the weirdest all-pass filter music applications of all are frequency shifters. The Barberpole Phaser uses a constant frequency shift resulting in an infinitely ascending or descending sweep, is perhaps the most exciting thing. It’s based on the auditory illusion of Shepard tones. Photo 49 shows one made by Harold Bode sold by Synth-Werk; Photo 49: a Barberpole phaser from Synth-Werk. 71 I will have to do one. You never know, it might lead to a new song, which has been the original drive for this electronic audio design path I’ve been on since 1979. The NE5532’s decline continued I was interested in why the all-pass filter seemed to be a distortion stress test for the op amps in our discussion of new vs old NE5532s last month. Ben Duncan suggested that the impedance of the all-pass stage may dip below 600Ω at some point. I measured it and found it was 940Ω at low frequencies, dropping to 600Ω above 8kHz. I needed to check this, since it is possible for the impedance to drop lower than expected with configurations based on the op amp differential amplifier. There is a situation where the resistor feeding the inverting input pin can have anti-phase signals on both ends, making the resistor appear lower than it is. I covered this phenomenon in the discrete differential buffer described in the April 2024 issue (see Fig.32 on page 64). I mentioned that the maximum output swing had been deleted from the datasheet for the new version of the Texas Instruments NE5532. An omission is always a bad sign, and it turns out my suspicion was valid. The output swing has dropped by a volt. Measured using ±15V rails with no load, the new device gave 27V peak-to-peak, or 22.5V peak-topeak into 600Ω. There was a touch of oscillation on the negative rail at clip as well. Using the old NE5532, it was 28V and 23.5V, respectively, with clean clipping. As the Editor wrote in last month’s editorial, the new design changes the input transistors from NPN to PNP. This reverses the direction of the DC bias currents. This means that careful orientation of electrolytic capacitors to agree with the resulting small Parts List – Linear-phase crossover 1 double-sided PCB coded AO-AUG-26, 108 × 75mm 3 NE5532 dual low-noise op amps, DIP-8 (IC1-IC3) 3 2-pin, 0.1”/2.54mm Molex-type polarised headers, housings and pins (J1, J2 & J4) 1 3-pin, 0.1”/2.54mm Molex-type polarised header, housing and pins (J3) 1 4-pin, 0.1”/2.54mm Molex-type polarised header, housing and pins (J5) Capacitors 8 100nF close-tolerance frequency-determining capacitors* (C1-C8) 2 100µF 25V radial electrolytic (non-polarised preferred) (C11 & C13) 2 10-100μF 25V electrolytic (C14 & C15) 3 100nF 50V X7R ceramic, 5mm pitch (C16, C17 & C18) 3 47pF C0G/NP0 ceramic, 2.5mm pitch (C9, C10 & C12) * select the appropriate value from Table 1. I used 100nF ±2% RIFA polystyrene (see AOShop). Polypropylene and 250V high-voltage polyester will also work well. All should be matched within 2% of each other. Resistors (all ¼W axial thin-film ±1%) 1 4.3kΩ (E24) or 4.42kΩ (E96) (R7)** 5 2.2kΩ (R1-R5) 2 2kΩ (R6 & R9) 3 1.2kΩ (R14, R15 & R16) 4 1kΩ (R8, R10-R12) 2 47Ω (R17 & R18) ** two 2.2kΩ (E12) resistors in series gives closer to the optimum 4.4kΩ negative offsets introduced would now have to be reversed. This was a technique we employed at Calrec in Hebden Bridge, which I don’t think made any audible difference, but it may have made the capacitors last longer. Also, all those fancy mic-pre and phono circuits that paralleled extra input transistors with the internal transistors for lower noise won’t work with these new devices. Two good things (silver linings?) are that the bias current with the new design is half that of the old version, and the noise level also measures the same when used in a standard RIAA preamplifier circuit. These things are a clue as to the identity of the ‘new design’. I suspect it is a re-engineered version of the RC4558. There was a special low-noise (4nV/√Hz) version of this, the obsolete JRC NJM2041 used in Philips/Marantz amplifiers and Fostex tape machines. It was specified to run from ±22V rails maximum, so I suspect it was an RC4558 made on the same process as the NE5532. Conclusion That’s it regarding all-pass filters, at least for now. I recently attended Retrotech and I will have a report on that shortly, including some photos of interesting devices. I am increasingly yearning for retro technology, if only because devices made in the past were intended to be repairable. Many products made since about 2015 (including cars and washing machines) are intended to be disposable – use it until it breaks, then throw it away and buy a new PE one. Not me. Photo 50: the front panel of the finished all-pass filter we built last month. 72 Practical Electronics | August | 2026