Fullscreen HTML5Med Res (??MB)HTML4

AUDIO OUT AUDIO OUT L R By Jake Rothman Absolutely phased out (all-pass filters, part one) I was listening to 10cc’s studio masterpiece from 1975, I’m Not in Love, and the line “It’s just a silly phase I’m going through” made me think it’s about time I clarified all this ‘phase business’ that pervades audio. There’s too much ambiguity about the word phase, which is often conflated with polarity, which is about positive and negative parts of the waveform. An inverter or swapping the wires in a balanced line or loudspeaker feed can be used to reverse polarity, making the waveform upside-down, as shown in Fig.1. Unfortunately, many people incorrectly call signal inversion “flipping the phase”, “reversing the phase” or “out of phase”. Absolute polarity, also sometimes called absolute phase, is a voltage or pressure effect where a posi- tive-going signal event comes out as a positive-going voltage. Phase shift is really a function of time, a time delay that shifts waveforms relative to each other, as illustrated in Fig.2. It can also shift the relative cycles of different frequencies, such as harmonics, often making the waveform more symmetrical, as shown in Fig.3. In audio, we should use phase to denote only a time-based alignment or delay, but it is notoriously difficult to shift culturally ingrained linguistic errors. I was once told to relabel a button on an active crossover from “invert” to “phase Ø” on the basis it was what the customer expected. The Latin capital letter O or Ø is often used to denote invert on mixer channels (Photo 1) because there is little panel space, which I think is acceptable. Voltage or pressure Editor’s note: delaying a sinewave (or any wave+1V form symmetrical about Ch 2 0V) by half a cycle has the Time 0V same effect on the shape as inverting it, hence voltage inversion being referred –1V Polarity flip to as 180° phase shift or +1V inversion. However, that does not apply for asymCh 1 metric waveforms. Time 0V Also, besides 180° and integer multiples of it, all –1V other phase shifts can only Fig.1: inverting the polarity of a waveform flips it about the time axis. + be reliably created by delaying the signal (or by frequency-dependent networks that are equivalent to delays), as implied above. Get it right It is the duty of the audio engineer to ensure his equipment does not invert the polarity, since this can cause problems later on. This is important because many signal processing circuits necessarily invert the voltage, which can cause cancellations when effects are combined in the studio. The famous Quad 303 and 405 amplifiers inverted the signal. Many CD players were also inverting because the R/2R digital-­ to-analog converter (DAC) fed an inverting amplifier. There is even a polarity-corrected CD available of XTC’s Skylarking album because it was thought this was why the first CD release didn’t sound as good as the original vinyl. + – – + + – – Voltage +30° phase shift blue lagging red +1V 0V Phase Photo 1: the polarity inversion button on a mixer channel strip. All-pass filter delays fundamental which aligns with harmonic Fundamental Fundamental 2nd harmonic 90° out of phase (half level) +1V 2nd harmonic (half level) Asymmetric waveform 0V causes clipping on lower cycles Symmetric waveform –1V 90° 180° 270° 360° 450° 540° 630° 720° Fig.2: phase shift displaces the waveform in time. Practical Electronics | April | 2026 –1V Clipping Fig.3: frequency-dependent phase shift can change the waveshape. 69 on the 17 th of August 2025 at age 87, and +V yes, I did think about interviewing him! Time 0V H o w e v e r, w i t h complex multi-­t rack mixes, the polarity –V is likely to be pretty +V random. In fact, a good mixing engineer will Time ensure it is as random 0V as possible to get the loudest mix. Having –V occasional coincident peaks or an asymmetInverted ‘pip’ appears at bottom Fig.4: asymmetric test waveforms are rical mix with more Photo 2: checking energy on the positive speaker polarity useful for detecting polarity reversals or negative side wastes using a battery. and phase shifts. headroom. Vinyl playback systems usually Australian mixing engineer to the positive on the battery is have correct polarity. People seem Michael Paul Stavrou, in his book marked “+” or colour-coded red. to vary in their ability to detect Mixing With Your Mind, described The positive connection denotes absolute polarity. Personally, if this approach as “maximum il- positive polarity. With microI polarity-flip most music, I can- lusion/minimum voltage”. He phones, a gentle puff of air like not hear a difference, but I feel would change the polarity of some saying a plosive word, such as reassured if it’s right. This can instruments (p193) to prevent “pop” or “bus”, should give a posbe easily checked with a scope or peaks building up in the mix, for itive-going voltage for the positive digital audio workstation (DAW) instance, making the bass line the polarity condition. using solo instruments with pos- opposite polarity to the kick drum. One engineer’s polarity test was itively biased waveforms such as When listening to one’s own to burst a crisp packet right near kick drums and trumpets. voice in headphones while singing my ear while I was miking up the Ted Fletcher of tfpro, an au- in the studio or broadcasting, it’s drum kit. Very annoying! dio engineer from whom I have essential that the absolute polarity It’s a rare thing in life where learned a lot, is convinced kick is correct. Very strange effects can two wrongs make a right, but if drums sound better if the absolute occur if cancellation occurs with the polarity is wrong, it can be polarity is correct. Ted sadly died bone conduction through one’s fixed by an inversion somewhere head or acoustic leakage else in the chain. In a stereo sys1:1 transformer through the ear pads. tem, it’s fine if both speakers are Finish Start Input Output It can be difficult to wired with the wrong polarity, but check the polarity of a a horrid hole-in-the-middle effect Primary Secondary winding winding simple sinewave through and bass loss is the result if only Start Finish a circuit, unless one uses one side is inverted. 0V 0V a dual-trace oscilloscope. This used to be one of the bigOne trace needs to be gest problems when I sold Hi-Fi +25V connected to the output systems. It didn’t help that the 100pF of the signal generator and speaker twin cable we supplied 39kΩ 300Ω 6.2kΩ 100µF triggered from it, the other was clear PVC insulated, printed Output to the probe. with very pale “+” signs along one 100Ω BD140 Inverter output Another way to do it side that nobody over 40 could see. load 2kΩ minimum with a single-trace scope is to use an asymmetrical The inverter Input 2.2µF The simplest audio inverter is a test waveform, such as the BC550C ones illustrated in Fig.4. A transformer, but it’s cheaper (and distorted sinewave with introduces less distortion) to do it 16kΩ 330Ω even harmonics should with an op amp, as shown in Fig.5. A switchable inverter can be do it, such as a full0V 0V wave rectified one. These made by using a switch to either 47pF waveforms are also good feed the signal through or bypass for showing frequency-­ an inverter, but there is a much 10kΩ better circuit that only uses a sindependent phase shifts. gle-pole switch, shown in Fig.6. I With loudspeakers, it’s 10µF Input 100kΩ V+ 2 simply a matter of con- suspect this evolved from differ100Ω 100µF – 7 6 Output NE5534 necting a low-voltage (3- ential op amp circuits – I don’t 3 4 + 6V) battery and seeing if know who invented it. 8 5 With the switch in the open the diaphragm moves out, as shown in Photo 2. If position, it is a buffer/follower, V– 10pF the cone jumps outwards and when the switch is closed, 0V 0V Fig.5: classic audio inverter circuits: transformer when the battery is con- it becomes an inverter, with the based, common-emitter and op amp versions. nected, the terminal going resistor in parallel with the input. + ‘Pip’ added to sinewave + + + 70 Practical Electronics | April | 2026 10µF 3 10kΩ 100kΩ (DC path) V+ – NE5534 7 6 Input 100Ω 100µF VR1 10kΩ 5 8 3 V+ – NE5534 V– 5 22pF 0V 0V 0V 7 6 4 + S1 22pF Output rotating VR1 CW 2 CW Output 4 + 10kΩ 10µF + 2 + Input 10kΩ 10kΩ + 10kΩ 47pF 47pF 100Ω 100µF + Fig.6: this switchable inverter uses only a single-pole switch. Output 8 Middle V– 0V Inverted Fig.7: the ‘attenuverter’ gives a range of gains from -1 (inverting amplifier) through zero to +1 (unity-gain buffer). The ‘attenuverter’ correct phase becomes critical. each other. This variation of group I thought I knew every analog Otherwise, unwanted cancellation delay with frequency is a type of audio circuit block in existence, notches occur in the frequency re- time distortion, with an ideal audio but Grindle brought the term ‘at- sponse. The main causes of phase system having a constant, short tenuverter’ to my attention. It is shift in audio are filters and time delay across the whole range. basically a level control (shown delays, both acoustic and those Rapidly increasing group delay in Fig.7) that has zero output in from digital processing (latency). as the frequency decreases, as exthe middle of its rotation, gives an It was believed until the mid- hibited by bass reflex enclosures inverted output on the anti-clock- 1980s that phase shifts were gener- and multiple coupling capacitors, wise side and an uninverted output ally inaudible, but improved source is especially noticeable, manifestwhen turned clockwise. quality has enabled the effect of ing itself as booming and lagging It is very useful for mixing two scrambled phase to be heard. This bass. things together and for synth con- is a subtle effect, though; it is not trol voltages. The first published at- as noticeable as frequency response The all-pass filter tenuverter circuit I could find was errors. It generally manifests itself Finally, we get onto the all-singby Ben Sullivan in the Circuit Ideas as a smearing of transients and a ing, all-dancing circuit that can section of Electronics and Wireless lack of clarity or mushiness. shift phase while having a flat World, August 1994 (p688). Timing is also affected, with frequency response. It is called parts of the frequency spectrum an all-pass filter for this reason. sounding detached. It appears the It is surprisingly simple to make Phase vs waveform shape Like polarity, phase is mostly ear is more sensitive to phase errors one by swapping the switch in the discernible in a relative context. at the lower end, since the time switchable inverter for a capacitor, You can easily hear when one shifts are longer. Phase is measured as shown in Fig.8. speaker in a stereo pair is invert- in cycles, so longer cycles have a If one thinks of a capacitor as a ed, but if both are inverted it may longer period for the phase shifts short circuit at high frequencies go unnoticed in perpetuity. With to be relative to, giving a more and an open circuit at low frequenphase, a single frequency (a sine- conspicuous effect. cies, it can be seen how the circuit This is often heard on the bass transitions from non-inverting to wave) at any phase has an identical where subwoofers with high-order inverting as the frequency increaswaveform. Take a harmonically complex roll-offs and crossovers are used. es, as demonstrated in Fig.9. wave and change the phase of those At the higher frequency end, it’s The interesting bit is in the midharmonics relative to each other, more subtle, mainly noticeable dle, where the slope of the phase and the overall shape may change with MP3 compression (although change is most rapid at 90° where the dramatically, but it will sound not as much as the bandwidth resistor and capacitor’s reactances the same unless the phase shift is limitation), cassette tapes and old are equal. This is the turnover freextreme and changing. I’ve made CD players and DACs with steep quency, and it is given by the standa video on this. It is the frequen- output recovery filters. ard formula 1 ÷ 2πRC. The phase θ cy amplitude detection model of (Greek letter theta) is measured in human hearing, which works like Group delay degrees and is given by 2tan-1(1 ÷ The cause of these audible phase 2πRC); it is 90° at this point. a spectrum analyser, that’s responThe time delay is given by t = sible for this psychoacoustic effect. errors is a variation in delay with frequency. This problem is often 2RC. I expected this 90° point to be Phase errors described by sound engineers as where the group delay is highest If different sources are combined, group delay, but the term group in accordance with the maths, but such as in multi-way speakers, delay is strictly a time delay. Phase change Mathematically, it 47pF 0° is derived from the 10kΩ rate of change of phase with respect 10kΩ V+ 2 to frequency. –90° 100µF Input 10µF 100Ω – 7 6 In audio systems, Output NE5534 3 4 + group delay often 8 5 10kΩ varies with frequen- –180° 4nF 100kΩ cy, so that a ‘group’ 20Hz 4kHz 20kHz (DC path) V– 22pF Frequency (logarithmic) of different frequen0V 0V fc = 4kHz cy signals will arrive Fig.9: the polarity/phase change of the at different times to all-pass filter in Fig.8 vs frequency. Fig.8: a single-stage all-pass filter circuit. + Practical Electronics | April | 2026 + S1 closed: inverter S1 open: follower (non-inverting) 71 Time (µs) Multiple-order all-pass filters with R3 10kΩ greater phase shift R2 and time delay can 10kΩ V+ R4 2 be made by cas100µF Input 10µF – 100Ω 7 6 cading the simple Output NE5534 3 4 + first-order circuit. If 8 5 4nF you put two in seR1 ries, it goes through 10kΩ V– 22pF 180° at the main 0V 0V inflection point and amounts, distorting the waveform finishes up at 360°, a complete ro(see Photos 3 & 4). To reveal the tation, as shown in Fig.12. frequency-dependent delay (group In Fig.13, four stages are cascaddelay), the measurement must com- ed for a tweeter delay; this seems pare the timing of different frequen- to be the most popular approach. cy components, for example using This gives a 720° phase shift and filtered or band-limited signals. 4 × 27.6µs delays for a total of Finally, the time delay for a 110µs. That is sufficient to comspecific frequency and given R and pensate for a 38mm path differC is t = 2RC ÷ (1 + [2πfRC]2). For ence. These differences often come example, Barry Porter’s Active-8 about either because the radiating loudspeaker in ETI, October 1984 part of one speaker driver is more crosses over at 3.1kHz (a typical recessed than another due to the two-way loudspeaker crossover shape of the speaker cones, or from frequency) with R = 10kΩ and C = driver mounting practicalities like 1.5nF, giving t = 27.6µs. flanges and such (often both). The position of the capacitor and Second-order all-pass filters can resistor can be reversed if desired be made around a single op amp, but (see Fig.11) and this will make the the increased noise and distortion phase change go the other way, is not worth it. There is another which may be useful to avoid an design we’ll look at next month that inverting stage. The high-frequen- uses two op amps, giving no cost cy distortion is worse with this saving, but the Q can be higher and arrangement due to the increased the time delay frequency range is common-mode voltage across the more consistent than two cascaded op amp’s input terminals. single stages. The values of resistors R2 and R3 High-Q phase changes can sound should be made as low as possible quite ‘ringy’; could this be a useful to minimise Johnson noise. I use studio effect? 1.3kΩ, the lowest allowable with There is also a discrete version, NE5532s. It is assumed that the cir- shown in Fig.14, which was used cuit is driven from a low-impedance in the B&W Active 1 loudspeaker, source, such as a preceding buffer. giving reduced noise. Fig.11: a reversed CR all-pass filter. + 40 47pF + 80 20 0 1kHz 10kHz Frequency (logarithmic) 20kHz Fig.10: the group delay of the allpass filter in Fig.8. Phase change 0° –180° fc = 1/2πRC –360° 20Hz 4kHz Frequency (logarithmic) 20kHz Fig.12: a complete phase rotation by cascading two all-pass stages. it’s not in some books. The curve, shown in Fig.10, has a shape more like a low-pass filter. An article in the Journal of the Audio Engineering Society, December 1988, by Dene Jensen (who is an expert on group delay in audio transformers) shows what I would expect to see: a peak in group delay at the point of maximum phase shift. I expect ambiguity with audio gurus, but not in maths. The delay introduced by an allpass filter is not constant with frequency. Measuring it using a square wave can be misleading, because the harmonics that make up the square wave are shifted by different 100pF 100pF 100pF 100pF 10kΩ 10kΩ 10kΩ 10kΩ 1.3kΩ 1.3kΩ Input Input from low-impedance 1.3kΩ source 12nF 1.3kΩ 1.3kΩ – – – – NE5532 NE5532 NE5532 NE5532 + + + + 1.3kΩ 1.3kΩ 12nF Output 1.3kΩ 12nF 12nF fc = 10.2kHz 0V 0V Fig.13: all-pass filter stages can be cascaded for a longer delay in applications like driver alignment compensation. Two-stage all-pass filter 6.54nF Buffer +18V 3.3kΩ 3.3kΩ 5.6kΩ 4.7nF 50mH centre-tapped Input Output CT 1.5nF BC550 60µs delay at low frequency BC550 1.5nF Input BC556 24kΩ 24kΩ 100nF BC550 22kΩ 0V 3.3kΩ High-frequency Output 3.3kΩ 72nF 0V –18V Fig.14: the discrete tweeter delay circuit used in the B&W Active 1. 72 Load = 833Ω 4.55mH 0V Fig.15: a tape group delay equaliser for the Revox A700. Practical Electronics | April | 2026 Photo 3: phase shifting a sinewave – no shape change. Photo 4: on the other hand, a square wave... Lattice all-pass filter Fig.16: a passive tweeter delay for a small 0.18mH bookshelf 4.7µF speaker. + 5.6Ω* + 0.18mH 4.7µF 680nF* Input from highpass filter *Zobel network Morel MDT29 dome tweeter Crossover for modified LS3/5A speaker 6.8Ω 2.5W High-pass filter 4.7µF Both units flush mounted centre distance on baffle = 120mm 4.3µF Input Output 0.27mH 6.8Ω To lattice filter Photo 6: a traditional lab phase meter. Passive all-pass filters In January 1981, a design to straighten 0V out the group delay of 6.8µF* a Revox tape recorder *Adjust to suit was published in Hi-Fi B110A used 56Ω* values for old KEFs News and Record Re10W view. This used a “T” network all-pass filter and is shown in Fig.15. 2.7mH + It employs complex KEF/Falcon 3µF B110A pot-core inductors and odd-value capacitors. Ampex ATR 100 multi-­ track machines had phase correction built in. The lattice filter is another interesting topology used for passive loudspeakers and old-fashioned studio phase processors. It is illustrated in Fig.16. 1µF 0V Low-pass filter 82Ω 2W Input 1.5mH 3.7µF 0V Measuring phase Photo 3 shows phase shifts on an oscilloscope, first with sinewaves and then (Photo 4) on square waves. Photo 5: a studio phase correlation meter. Practical Electronics | April | 2026 You can see how the shape of the sinewave is preserved but the square wave’s shape is definitely not! An important and useful tool in audio mixing is the phase correlation meter shown in Photo 5, which compares the phase difference between the left and right channels independent of amplitude. When it is in the middle, that is normal stereo; when fully deflected, it is mono; and when deflected backwards, it indicates a 180° phase shift or reversed polarity. These meters were very useful for stereo tape head azimuth alignment using a mono test tape. Photo 6 shows a traditional phase meter used in an electronics lab, which is not so useful for audio work, since it does not have a centre zero. Lissajous figures are a classic way to show phase relationships by applying the two signals to be compared using the X and Y inputs of an oscilloscope. They can be an 73 Photo 7: the result of phase-shifting a wonky ramp. art form in their own right, as seen in Photos 7 & 8. Phasers The phaser effect is possibly the most common use for allpass filters in audio, which is not surprising as it is one of the best sound effects around. Jean-Michel Jarre’s groundbreaking 1976 synth hit album Oxygène was virtually built around it. I designed and built my own dual-channel studio quality rack mount phaser, visible in Photo 9. This used a total of 12 all-pass filter stages in each channel. Phasers produce their wonderful spacey effect by sweeping several notches up and down across the audio spectrum. The notches are created through phase cancellation. To achieve this, the direct signal is mixed with an all-pass filtered version. Two all-pass filter stages are needed for each notch, giving the required 180º for a cancellation notch with the original signal. This element can form the basis of a comb filter. True comb filtering used for flanging has many harmonically related narrow notches; it needs a real delay line made using digital or bucket-brigade techniques (or two tape machines). For a nice-sounding phaser, three notches are required, so six stages in total. A six-gang pot could be used to sweep the notches, but it would be an expensive component. A better solution is to use voltage control. Voltage-controlled all-pass filters Fig.17 shows all the filter control techniques I have found effective. The pot sections could be replaced directly by optoisolated JFETs or Vactrols (the classic LED + LDR combination), both of which are expensive, but they offer total control isolation. 1.3kΩ 1.3kΩ – Output 4.7nF 0V 0V – Output + 130Ω 1.3kΩ Control Vin Input 74 Output Current out – + 4.7nF Inverter 0V 0V + 2150 JFET BF244 ½ LM13700 (transconductance amplifier) 0V Current (I) control 1.3kΩ + 4.7nF 1.3kΩ Output Input + 1.3kΩ – – Current out 0V 10kΩ 10kΩ 1.2kΩ 4.7nF 0V Fig.17: several voltage control methods for all-pass filters. 1.3kΩ Current (I) control Vactrol LED input Control input >5mA needed 0V 1.3kΩ – + Output 620Ω 4.7nF 0V ÷10 attenuator Output + + Input – Input Opto-FET Input – 100kΩ Anti-log A low-cost voltage-controlled resistor is the good old JFET, which most phasers use, including the Colorsound Phazex shown in Photo 10 and Fig.18. To apply these, their gate control voltage needs to be referenced to ground, so it is necessary to reverse the position of the capacitor in the filter relative to the resistor. The JFETs have to be selected so that their pinch-off (Vp) voltages match to ensure that the filters track each other. This is a major problem with this circuit, since V p can vary over 50% between devices. I have found the cheap Toshiba 2SK2145 SMT dual JFETs help with their better matching. When we made phasers at Colorsound, I had to make a special JFET measuring jig (described in an earlier issue of this magazine, back when it was called Everyday Practical Electronics). There are two chips that can be used in a way 1.3kΩ 1.3kΩ Input Photo 8: Lissajous figures based on that wonky ramp. Blackmer VCA 0V 0V 0V Practical Electronics | April | 2026 Photo 9: a special 12-stage studio phaser I made. R4 10kΩ R3 47kΩ S1a R1 47kΩ C1 47nF 6 5 6 – IC1b TL062 + 7 5 4 0V 0V + IC2b TL062 7 3 + C2 47nF R5 22kΩ R2 470kΩ 2 – C3 47nF R8 22kΩ TR1 2N5457 TP1 100kΩ 0V C8 10nF 3 IC2a TL062 4 0V TR2 2N5457 1 C4 47nF R11 22kΩ Mixer summing amplifier R13 10kΩ 3 C5 47nF R14 22kΩ – IC3a TL062 + Dip Freq (Hz) R15 10kΩ 2 – IC3b 7 TL062 5 + 4 0V TR3 2N5457 R + Direct R17 10kΩ 1 4 0V TR4 2N5457 VR2 10kΩ log C6 2.2µF 50V S1b R18 10kΩ R16 10kΩ R19 150kΩ Output S2 Resonance 0V 0V R20 1.5MΩ – IC1a TL062 TP2 1MΩ Sweep + C11 47nF C10 10nF R24 150kΩ 1 R26 6.8kΩ R25 470kΩ Practical Electronics | April | 2026 9V BAT1 R21 47kΩ Speed R23 150kΩ 6 – + R12 10kΩ Freq (Hz) FET bias C7 33µF 10V 2 R10 10kΩ R9 10kΩ Peak + VR1 470kΩ Reverse log C9 10µF 10V + Input R7 10kΩ R6 10kΩ – of the right-hand channel. + Photo 10: the Colorsound Phazex. L Amplitude (dB) Next month Possibly the best information available on all-pass filters is in Douglas Self’s book, The Design of Active Crossovers (2nd Edition, 2018, ISBN 978 1 138 73303 9), which does a brilliant job of covering the electronic aspects. It does not cover combining the acoustic response of loudspeaker drivers with filters, though, which is still an experience-based ‘black art’. In part two, Direct signal we’ll look at loudspeakers and make an Input 6-stage all-pass filter all-pass filter Total phase shift = 1080° board with up to eight sections for multiple applications, such as a waveform ‘symmetriciser’, a multi-channel microphone phase adjuster and a tweeter time alignment tool, as well as some traditional phasers. As with capacitors, there seems to be much misleading information about phase in audio out there. Watch out for being made an April fool; I found an article on designing an “all-stop” filter by Ben Sullivan in Practical Wireless, April 1989. PE Left-hand channel has the inverse frequency response Fig.19 Amplitude (dB) Stereoising A phaser can be made pseudo-­ stereo by subtracting the all-passed signal from the direct signal on one side and adding on the other, as shown in Fig.19, giving two opposite frequency response channels for left and right. The effect vanishes when played in mono. This trick gives a useful ‘stereoising’ effect on instruments such as pianos and it does not necessarily need to be swept. – that avoids both the common mode distortion and matching problems, are the LM13700 transconductance amps, which are noisy, and Blackmer VCAs. These devices from That Corporation avoid the noise of transconductance amps, give very low distortion and are ideal for a studio-quality phaser. However, they have an exponential control characteristic, are inverting and are very expensive. Another chip type I’m interested in trying is the Dallas digitally controlled pot, but I suspect using those in the circuit would result in a ‘zipper’ noise. Fig.18: the Colorsound Phazex phaser circuit, © Macaris. This is the simplest possible circuit to obtain useable results. D1 1N4001 V+ Pins 8 (IC1 – IC3) C12 180pF R22 6.8kΩ 0V D2 4.7V 0V 0V 75