MXR Phase90: Cv ANALYSIS & DEPTH CONTROL
SOME BACKGROUND …
Being absolutely NO expert in historical anything my understanding is that Keith Barr created the MXR Phase90 and Phase45 in his garage around 1973 and put them on the market the following year. We can say that these two innovative phasor designs changed musicians’ perspective on pedals, as their sonic and physical qualities quickly became regarded as tops in the industry. Because of the sheer number of units sold they are among the best known of all the vintage-era phasors.
RELATIVE PERSPECTIVE …
To many players, the Shinei “Univibe” is considered the big daddy of Phasors – having come out in 1968 and associated with many psychedelic players such as Jimi Hendrix and David Gilmour, … as well, as the slightly less psychedelic Gerry Rafferty.
The Univibe is a bulky AC powered unit with a front-panel control for DEPTH and a RATE controller set by a foot-treadle mechanism similar to a volume or wah pedal … it has a strong character that is not always welcomed in a mix, especially outside of psychedelic music (why I really enjoy what Rafferty did with a vibe)
In comparison, the MXR Phase90 and Phase45 were much smaller, ran on a 9v battery exclusively and only had a RATE control on the front – with a fixed effect “strength/DEPTH” that was set by design …
the MXR units were known for sounding way more even and better-behaved – not so prone to making everybody sea-sick like some of the other phasor designs did. For this reason, not too many players seem to mind the compromised single-knob designs – the idea being, you simply had phasing or you didn’t … no fussing
Above all, the MXR Phasors filled a sonic niche not covered by the main alternative at the time, the Univibe … the MXR units provided “pro” level phasing along with some great noise specs, again improved further in the second generation Phase90
jFET BASED SIGNAL-PATH DISTORTION
On this page I won’t dwell much on the signal path design aspects other than simply state that the Phase45 operates two phasor stages with jFET biasing that includes a distortion cancellation scheme, while the Phase90 operates four stages without. Since we don’t see many other jFET phasor designs going to this trouble, new and old, we can say the MXR Phase45 is somewhat unique in this regard.
Still, in either case, and for that matter in any case involving jFET’s operating as Voltage-Controlled Resistors (VCR’s) in this way, the headroom of the signal-path is limited to around 100mVpp regardless of specific jFET’s used or the circuit’s specific operating voltages.
This headroom limitation may or may not be useful to a musician. For example, to someone like Donald Fagen who used a Phase90 on a Fender Rhodes the soft clipping of the jFET’s produces a very pleasing and characteristic sound. To a bass player, say, looking for clean phasing, these two pedals would likely not be a good first choice. So, jFET based phasors are associated with low-headoom. And, of course, we know in a highly distorted setting (EVH) that this low-headroom is not a problem, in fact possibly adding another dynamic compression element in an already complex sonic picture. There are no rules but rules to be broken here.
I will leave the analysis of Phase90 and Phase45 signal paths for another day and now focus solely on the design elements associated with the control side of the Phase90. To do that we need to first consider what questions and constraints Keith Barr must have faced when preparing the Phase90 circuit for mass manufacturing.
As a side note, for those looking towards a higher-headroom phasor of similar design should have a look at the MXR Phase100 which operates on opto-couplers … with a simple tweak it can be turned into a 4-stager like the Phase90 – basically yielding a high-headroom “optical” Phase90
MANUFACTURING CONSIDERATIONS
Design engineers not only have to come up with working solutions to a problem, but they have to make sure that whatever variance might crop up will not land their solution into some kind of no-go zone. The Phase45 and Phase90, being very similar in design approach, both face this kind of challenge – taking note of the fact that manufacture-able means keeping human interaction with each unit to a systematically set minimum. In the case of these two phasors we are talking about the adjustment of a single (internal) trimpot.
In order to arrive at this point a bunch of work needed to be done, and it is in this light that we can regard the Phase90 design as being a sophisticated juggling act existing between the three parts that always make up any Phasor effector circuit. These are: (i) the signal path, (ii) the controller circuit, and (iii) the interfacing elements that link the first two. Here we will focus on the last two aspects.
ABOUT jFET’s …
Don Morris, who took over the Phase90 design after Keith Barr left MXR, clearly stated in an interview that MXR had a special arrangement with a large semiconductor company to select “matched” quads of 2n5954 jFET devices for their Phase90. It is entirely possible that something else was involved (!) in the selection process …
Don examined dozens of failed units and searched for dual op-amps with low power consumption, and for alternate suppliers of the FETs.
After bread-boarding several versions, he laid out the new PCB and submitted the result for approval, which it received. Here’s what he did:Selected the TL062, the low-power version of the TL0xx line. It satisfied the power-reduction requirement and did not alter the sound in any significant way. And with a dual op-amp, he gained valuable real estate on the board!
Added simple single-pole pre-emphasis/de-emphasis stage to reduce unit hiss. This idea is not seen in the current-production models. Found a FET supplier that pre-screened and matched the FETs to a tighter spec. This reduced the failure rate to 3%. Now that the units accommodate AC adapters, most of any decent dual 8-pin op-amp can be used, and an on/off LED–a battery-killer in the past–is essential now. But that was the thinking and experience at that time.
So, there has always been lots of mention made about jFET matching and all that, but one thing that still remains a largely dark area in this design is what exactly happens on the control side of the equation; more so, how the two sides interact. To help understand how the two sides play with each other we need to understand a little bit about jFET “VCR” theory. The paper to read is the Vishay App Note:
VISHAY AN-105 APPLICATION NOTE
In a nutshell, the jFET VCR’s have their (input) Vgs voltages lying inside a portion of the Vgs(off)-to-zero range; Vgs(off) being a negative voltage for the type of devices (n-channel) used here. In this way we can talk about a normalized percentage of the control range being used, regardless of the actual (matched x4) Vgs(off) voltage.
>>> Part of the design juggling act here involves choosing a fixed design controller circuit (with one degree of variability) and cause enough of an input voltage variation in the FET’s to create an acceptable degree of phasing.
The question(s) then becomes, (i) how much of a control voltage span gets seen at the FET gates, and (ii) how does that translate into a percentage of resistance variation, which proportionately translates into phasing.
Because some players would like to know if this aspect of the Phase90’s response can be tweaked or even made a adjustable this opens up an interesting analysis and even a potential re-design challenge.
TRIPLE-PURPOSE ZENER DIODE
A central element in the Phase90 design is a voltage reference created by using a 10k resistor to bias a 1n4733 Zener diode from the battery. The Zener diode is listed as having a 5.1 volt breakdown voltage but being biased weakly at around 0.5mA the Zener’s voltage lies more around 4.8 volts (closer to mid-rail actually).
In its very moderate capacity to hold a steady voltage the Zener acts as (i) a low-ish impedance voltage reference to the Phasor’s signal path (the AC ground), (ii) a source for the trimpot-variable BIAS reference, and (iii) as a reference for the astable LFO circuit.
THE STOCK PHASE-90 LFO
To study things a little more in depth let’s start from the basics. At the heart of the Phase90 LFO is an Astable Multi-vibrator circuit, and at the heart of that is a Hysteretic op-amp Comparator. Some would call it a op-amp Schmidtt trigger also. Spice simulation (here via TINA-TI) can be used to illustrate the steps that lead to the full design.
For a single supply design we can sweep the input linearly between rail limits (gnd and 9v) yielding a triangle wave, a waveform that varies voltage linearly in time. The first plot shows the transition points of the hysteretic comparator …
Because the feedback is “positive” this leads to circuit behavior with only two stable point, each lying at the saturation extremes of the op-amp. In this circuit the output slams between these two extremes, whose values are set by the supply voltage and the saturation characteristics of the op-amp used. These figures that are available in manufacture datasheets and covered by device macro-modeling inside SPICE simulation software.
Because the “drop-out” voltage is about the same on the upper and lower sides of the op-amp’s output stage we end up with a square wave that is vertically symmetric, meaning that is will have an average value (first term in a Fourier expansion) that is very close to mid-rail. This is an observation that we will make use of shortly.
Now, if we add a timing circuit we can build an Astable Mutli-Vibrator, a circuit that keeps trying to run away on itself.
The voltage seen at the timing Capacitor can be regarded as a Low-Pass filtering of the output square wave, resulting in a pseudo-triangle waveform – again, with an average value that is close to mid-rail.
PHASE-90 CONTROL VOLTAGE MIXER
The Cv voltage that is seen directly at the FET gates is produced by a high-resistance voltage mixer linking the waveform present at the timing capacitor and the DC voltage present on the BIAS trimpot. The resistive mixer is comprised of a 1Meg-ohm resistor and a 3.9Meg-ohm one wired in series – the mid-point feeding all four FET gates at once. To be clear, the Vgs (VCR input) voltage is formed by the difference in voltage seen at the Gates and the Sources – the later being set directly by the Zener diode.
Because these two mixer resistors together form a voltage divider and because the lower impedance BIAS trimpot also acts acts as a relatively stable DC voltage reference (AC ground) the pseudo-triangle wave form seen at the charging capacitor invariably gets scaled down as a result along the way. A waveform that starts out at about 1.40 voltsPP ends up measuring about 0.30 voltsPP by the time it hits the FET gates. It is clear that this attenuation ratio also remains largely the same regardless of the BIAS trimpot setting, as the following simulation plots show:
100% rotation …
80% rotation …
20% rotation …
So, all this begs a question that I don’t see ever asked, that is: how much phasing range is produced in this arrangement (versus how much is necessary or wanted), and how does it vary against FET parameters like Vgs(off).
If we understand the ViSHAY paper we know that resistance variation in jFET VCR’s is linked to proportion of Vgs range used. So, if the 300mV control voltage span is applied to a jFET with a Vgs(off) of, say, -3.00volt versus a jFET with Vgs(off) of -1.00volt, then we’ll have 10% at most of the useful control range traversed in the first case versus 30% in the second. Surely, we would experience way more phasing effect in the later case.
TWEAKING THE LFO “DEPTH” RESPONSE & THE ROLE OF A DYING BATTERY
As I stated out saying in the beginning, the design is a real juggling act and this is certainly no criticism of the final solution adopted BY Keith Barr – quite the contrary.
Certainly, there are those – maybe in a DIY capacity – who would like to make things adjustable in the Phase90. So, let’s look at a popular approach to doing this and the associated drawbacks that come with.
This approach involves replacing the 1Meg ohm mixing resistor with a variable resistor. Notice, that if we sum the two resistors we end up with 4.9Meg ohm – very close to 5Meg ohm. So, we could in essence replace (conceptually or physically) these two resistors my a 5Meg ohm potentiometer. But simply altering the 1Meg ohm mixing resistor achieves a similar purpose.
Now, here’s the catch … with a little bit of thought it should become clear that two sides of the mixing circuit very likely operate at different DC (average) voltages, so that when we vary the 1Meg resistor on its won, for example, we end up altering the vertical BIAS picture, as shown in this set of simulation plots where the 1Meg resistsor is made to vary between 200k and 5Meg ohms:
This means that bias drift will accompany AC signal DEPTH variation when using this method – which explains why many players who try this mod like it, but only a degree (from reviews I’ve read). This makes sense. The good news is that with this tweak we can tailor the Cv voltage swing going to our jFETs to create a possibly better/stronger phasing effect, all we need to do is re-bias and we’re good.
Another thing that will cause relative bias drift is the dying of the battery – which is one reason why Don Morris put an AC adapter jacks on the units he re-designed. Because the Zener voltage doesn’t drop at the same rate as the mid-rail of the LFO circuit in a dying battery scenario, it is basically impossible to guarantee stable relative biasing of the FETs as it only takes a few tens of miLli-volts to take the biasing into a range of poor behavior. Still, all other aspects of performance (including sensitivity to a dying battery) are maintained when this tweak is done to an original unit.
ADVICE FOR BUILDERS OF ECONOMY “BATTERY-ONLY” CLONES
Since there are players who steadfastly refuse to have AC power adapters in their rig (I get it), they might want to build a battery-only unit that is DEPTH adjustable via the idea just presented. For those who prefer battery-only operation, I would advise to bring the BIAS as an external control and expect to have to tweak it when adjusting DEPTH. Hey, this is as old-school analogue as it gets!
SLIGHT RE-DESIGN FOR NON-BATTERY USERS
One thing we can do, as mentioned above, is replace the two mixing resistor by a 5Meg pot. If we do away with a varying battery voltage, and now align the BIAS reference voltage with the mid-RAIL average of the Timing capacitor waveform – which are now both fixed and adjusted to be equal – we can produce a DEPTH control that leaves the jFET biasing levels constant in average, and therefor biasing invariant.
The key to making this work, without incurring any weird kinks or blips in the scaled waveform, is by making sure the DC reference has a low drive impedance, which is done by buffering the BIAS trimpot by an active emitter-follower stage …
the next question, then, becomes how to BIAS the jFETs in turn. Simply, this is done creating a separate AC ground source that is vertically adjustable – something that is easily done with a trimpot and op-amp or transistor voltage follower.
At this point, the circuits begins to acquire minor amounts complexity; leaving behind the super-simple stock design, but with the advantage of having only two front panel controls: SPEED and DEPTH. Also, more importantly, the Cv signal swing seen at the gates can be made as large as 1.40vPP from its previous 0.30vPP.
A GENERALIZED APPROACH USING DUAL OP-AMP TRIANGLE WAVE GENERATOR
For those who are curious to experiment, we can produce a bone-fide triangle waveform through the usual dual op-amp means, and have its output scaled down with no difficulty. We should note that both dual op-amp and dual ota triangle wave oscillators generally need to live at mid-rail in order to produce a horizontally symmetric waveform, so we start with that as a first constraint.
Then, as before, an adjustable AC ground source can be used to provide a reference of the signal path where the jFET Source connections and therefore back-biasing takes place. In this regard, biasing and Cv signal scaling can be made independently from one another. To limit the range, a padding resistor is included between oscillator output and the DEPTH control.
click on image to see variational simulations …
