EA Tremolo revisited--rEAgenerated Tremolo

[Transmogrifox]

It seems a common complaint that the EA Tremolo leaves some performance to be desired.  It's a great tremolo, sounds very smooth, and is simple to build, but it just doesn't have enough depth for complete satisfaction.

EDIT: 
Here's the final result.  It has as much depth as the opto based tremolo in my amplifier, but the same smooth, sweet sound of the EA Tremolo.  I call it "rEAgenerated"
It also doubles as a nice sounding FET booster, and overdrives a little when you crank the input.

www.geocities.com/transmogrifox/rEAgenerated_Trem

Finally, I recommend to any of you to pick up a batch of FNJ598J FETs.  They're much cheaper than the commonly used FETs I have seen in pedals, and this use has shown me that they are by no means inferior for use in stompboxes.

/EDIT

I went to testing an idea on the breadboard this afternoon and came out with an improvement.  I added a constant current source to increase resistance seen by the source of the FET, thus reducing gain in the "off" mode for the JFET varable resistor.  I increased the gain of the cirucuit somewhat (though this largely depends whether you use a BJT, JFET, or MOSFET for the main amplifier), and put the level control on the front end.  By doing this, we reduce the input level, and compensate with the variable gain stage.  Because the expected input can be lower, we can amplify it more without having to worry about headroom.  This is more desirable because we can get a greater overall gain change.   It's easier to go up than down.

Here are my (approximate, hand-waving) calculations on it:
The BJT current source of that type generally has a resistance from Collector to Emitter of about 100k.  If the control FET is turned off completely, or removed, the equvalent gain is 6.8/100, which comes out to about -23dBV.

At maximum gain (if using a BJT) the number is 6.8/0.22, which comes to about 30dBV.

By using a BJT on the front end, I can assume a dynamic normalized gain of about 53dB. 

With a MOSFET, I estimate an equivalent FET source resistance at around 100 ohms.  The gain is then, 6.8/.320, which is around 26dBV.

We still see a near 50dB dynamic range.  This is a marked improvement, and the difference is certainly audible.  The tremolo improved a lot in my mind.  I ended up going with the JFET for the amplifier because it's not terribly noisey (like an N-Channel FET), and it doesn't have quite as brittle of a crack as the BJT when it's overdriven.

The numbers I threw out assumed that the control JFET max resistance is much higher than the BJT's Rce, and that it's much lower than th 220 ohm.  Both assumptions I know to be false, so I'm thinking the circuit achieves more like 34dB dynamic gain in the worst case (still not bad considering the current design probably gets about 20dB).

EDIT:
I was at the time of writing not 100% satisfied with what had been achieved at this point.  I have posted this version that did not totally satisfy me for those of you who are interested in viewing various steps in a development process:
/EDIT

I'm still not 100% satisfied, so I intend to add a resistor in the emitter of the current source to see if I can get the extinction considerably deeper.  Just can't do too much of that kind of thing or we start losing headroom.

Here's the link:
http://www.geocities.com/transmogrifox/EA_Revisited.JPG

Note the uncommon selection of JFET.  This seems to be a goodie.  I got them from Mouser.  They're much cheaper than the standard J201 or 2N5458, and have the same pinout as the 2N5089 so you can easily swap and plug-and-play between these, MOSFETs and various BJT's with one minor bias adjustment to the circuit.

EDIT:
I would like to add (see posts below) that based on the information given by brett, the J201 would be a good substitution to the FET shown.  Based on the voltages I measured coming off the LFO, I am almost certain that a 2N5458 as the control FET would work fine. 

[Processaurus]

Cool!  I've been intrigued by your idea addressing the depth issue with this circuit since you proposed it a while ago, its great you found time to put it together.  I've got to try this soon, I'm assuming the unusual FET is just your favorite, and that others will work (and improve the depth)?

The idea of the volume in front is also very good, the EA has so much gain that it distorts badly with hot pickups. 

[Transmogrifox]

I think the other FETs like the 2N5458 would work. There is one quality that the FJN598J have that the others don't, and this is a lower Vgs OFF of about 1.5V.  The others are at about 3 or 5 V (and like JFETs, varies a lot from part to part).

The LFO can theoretically output +/- 4.5 V to the gate, however you need to do some changing around the area near the depth pot to get it to do this. 

I'm going to do some more experimenting this afternoon.  I'm going to try the Widlar current mirror, since it  is able to multiply Rce by a significant factor such that it won't even be a part of the equation at all.  I'll also reconfigure the depth control to get more depth out of the FET

[brett]

Hi
I find that the J201 has a Vgs closer to zero than most other JFETs.  It's mostly between 1 and 2 volts.  And it tyically has high transconductance.
cheers

[Transmogrifox]

Ha! Perfection.  Here's the final verdict on my afternoon's experimenting:
http://www.geocities.com/transmogrifox/rEAgenerated_Tremolo

It's beautiful.  It has as much depth as the tremolo on my amp.  The Widlar Current Source take-off was the trick it needed.  The final cherry on top was to reconfigure the circuit around the depth control.  It sounds absolutely incredible and I like using it as a booster, too.  All you have to do is put in a switch to short the LFO output, and you're set.  I'll try to get some clips recorded next chance I get.

Hi
I find that the J201 has a Vgs closer to zero than most other JFETs.  It's mostly between 1 and 2 volts.  And it tyically has high transconductance.
cheers

Thanks for the info.  In that case, the J201 would make a fine alternative to the FNJ598J.  Might make for a bit more distortable booster as well.  The way I have this circuit configured now, I can't think of any reason a 2N5458 wouldn't work as the control FET.  I wouldn't recommend using a 5458 for the amplifier, though since it WILL limit the headroom quite a bit.

This FNJ598J leaves the source at about 1.5Volts, while the drain can be biased to about 6V.


With these modifications, you can still use the layout for the original EA Tremolo, but there will be some added floating components to make it work.  That's said just to encourage you who already have an EA (and wish for more depth and less distortion) can mod it without having to scrap the entire circuit.

A 500k pot is recommended on the input, though I show a 250k pot on the schematic because that's what I was using on by breadboard, and it worked just fine.

[vhbram]

Hey,

I always wanted to make the EA Tremolo but everywhere I hear there is to less depth and other troubles, but I think I found the solution here.  I'm thinking of building your version but with the J201's.  Is there also more speed here and what does the "Double Trem Mod" actually do?  Some audio sample from your tremolo would be very welcome!

Thanks a lot,

Bram

[Transmogrifox]

I have been struggling with the toner transfer thing to make a PCB.  Finally on Sunday I got the PCB etched and thus disassebled it from my breadboard.  The board is populated and ready to add pots and switches...
All that to say, I'm hopefully close to being able to do audio samples. 

I made this one with the double trem option (and had it on my breadboard also).  It sounds like having one tremolo fed into another, both going at different rates.

It produces periodic rythmic patterns such as tripplets, or a sequenced sound.  It's pretty fun to play with and sounds good into a wah pedal.

The simple version doesn't address the speed problem.  The double trem mod gets around it because you can switch to either the high-speed oscillator or the lower speed oscillator.  The high-speed oscillator gets pretty fast, but doesn't go very slow.  The low-speed oscillator goes slow enough, but doesn't get very fast.  Between the two you can get about any rate you would want from a tremolo and combine the two for even more weirdness.

[vhbram]

Hey,

thanks a lot! I think this will be my next project.  Could you also add your PCB's (trem and double trem) to the schematics, that will help me a lot...

Greetings from Belgium,

Bram

[Transmogrifox]

Yes, I plan on putting a PDF project together.  It is layed out to fit into Hammond 1590BB sized box.  I added onboard relay switching with microcontroller and voltage regulator, so it took up a little more space.  I'll add some notes about how to do switching with a regular DPDT, or an SPDT and relay so you aren't stuck using a uC.

[vhbram]

Ok thanks I can't wait to see de pdf project (and some nice sound samples),

greetz,

Bram

[R.G.]

Great work there!

Only one suggestion - you could servo the JFET bias by adding an opamp to moniitor the drain voltgage and adjust the bipolar transistor bias with the output of the opamp. That would remove the bias adjustment.

Of course, with an onboard uC, you could do the same thing in code, looking at the heavily-averaged value of the drain voltage and then diddling the bias on the bipolar with that as well. A digital pot might work, but you could use resistors too.

[Transmogrifox]

Only one suggestion - you could servo the JFET bias by adding an opamp to moniitor the drain voltgage and adjust the bipolar transistor bias with the output of the opamp. That would remove the bias adjustment

You're just trying to improve the coolness factor 
Thanks

As I was reading that, it occured to me there's an even more simple way to do things here.  With the 1k resistor in the emitter off the CCS, the .7V drop becomes a good approximation, and I could probably just calculate it and specify a fixed value of resistor that requires no trimming.  I know why I had the pot--it's carried over from my first attempt where I wasn't using the pseudo-widlar current source and needed to bias the transistor more precisely.  The DC drain current is independent of the FET, thus the improvement in maximum depth.

Let's see... A 6.8k resistor needs a 3V drop, thus a current of 441uA.  441uA * 1k = 441mV drop.  441mV + 700mV = 1.14V.

Beta~400, so the base current is about 441uA/400 = 1.103uA. 

We'll arbitrarily put a 10k to ground.  This 10k has the above 1.14V drop across it, and I=114uA
114+1.103~115uA
So we need a resistor that creates a (9-1.14)V drop at 115uA = 68.2k

Wonderful.  Using a 5% 68k resistor, we don't have to do any more tweaking.

Instead of the trimmer, I shall replace the CCS bias with a 68K + 10k resistor divider.

[lovekraft0]

...you could servo the JFET bias by adding an opamp to moniitor the drain voltgage and adjust the bipolar transistor bias with the output of the opamp...

As I was reading that, it occured to me there's an even more simple way to do things here...

OK, so there must be at least two reliable ways to make JFET bias more or less device-independent. If either if you well-educated gents would be so kind as to explain the basic principle at work here in terms that a semi-literate guy with a bad liberal arts education and a breadboard background can understand, I for one would be forever even further in your debt - I read the Siliconix JFET biasing App Note, but I can't quite see how to apply it to real-world circuits. Any help would be greatly appreciated!

[Transmogrifox]

In this case, the JFET biasing is irrelevant to the underlying principle.  Whether the front-end amplifier transistor was a BJT, or a MOSFET, or in this case a JFET, we have the same principle at work for us.

For the moment, forget about the gain control (the capacitor, resistor, "variable resistor" FET).  Without that, this configuration is theoretically silent.

For the amplifier FET--think of it as a water valve.

The BJT in the schematic is a constant current source.  Think of it as being a water pump.  It's a pump that makes sure that there is always the same flow no matter what.  If the valve that supplies water (JFET) is open to a trickle, then the pump will build up a lot of pressure to get the same amount of water to flow (voltage being compared to pressure).  If the valve is wide open, then there is need for very little pressure to get the same amount of water to flow. 

Basically what this means is that the bias on the BJT sets the amount of water flow it demands from the colletor (pardon me while I butcher the technical details).  The FET is a valve that can be more open or closed.  In this circuit (ignoring the gain control network) the voltage on the gate of this FET "valve" doesn't matter, because the same flow is force through it no matter how hard it wants to restrict it.  A guitar signal would operate the valve to turn it more on & off, but as long as the high-impedance BJT constant current sink is driving the source-drain current, it doesn't matter what the guitar signal does.  It will not be reflected on the drain resistor.  This is why we are able to make it a deeper tremolo, because the depth of the tremolo is 99.999% dependent on the gain control network, and no longer on the biasing network.

Once you add the gain control, then things change.  As the guitar signal drives the gate of the FET more positive, it begins to charge the capacitor, which requires more  drain-source current, thus the voltage on the source resistor goes more negative.  When the gate voltage goes more negative, the "valve" begins to shut off.  But now, the constant current source has a reserve tank (capacitor) it can draw from without having to force flow from the valve.  Thus the drain voltage can go more positive.

As the control FET is turned off, then the impedance into the "reserve tank" looks very high (so high it's as if it wasn't there), so you're back to essentially the DC state where there is no gain on the signal seen at the gate of the FET.

About biasing JFETs with a resistor...I'll continue more later.  I have to go now.

[Transmogrifox]

OK, so there must be at least two reliable ways to make JFET bias more or less device-independent.

1.  Feedback
2.  Constant current source (as used in this circuit).

The constant current source is a form of a feedback network.  It's similar to sticking a very large valued resistor in the source.  The + about the CCS is that you can bias the JFET at a current where you can get appreciable transconductance.  The downside is that you have to add more parts into the circuit to do it.  It's still a lower parts count than a discrete component JFET op-amp.  In addition, the CCS allows you to get appreciable gains to get distortion, and doesn't "linearize" the JFET's transfer characteristic so you still have the nice clipping characteristic.  That's why I chose to use the JFET in this circuit.  The BJT was to brittle when it started to move toward distortion, and the MOSFET is to noisey for my taste.

I'm uncertain if I fully answered your question, but I hope I gave you enough thoughts to know what to ask next.  Who knows, maybe RG or Mark Hammer would be kind enough to jump in.  I know RG has posted a lot of this kind of information at GEO, and you could probably have a lot of questions answered there. 

[lovekraft0]

Thanks, that does help more than a little - let me see if I've got this close to correct. Please bear with me.  The concept here is to set the DC operating point of the JFET to a specific voltage by adjusting the current at it's source, right? Since the drain current is going to be the same as the source current, the voltage drop of the drain resistor (which is the DC operating point) can be controlled by controlling the source current, so for instance if you had a 10K drain resistor and wanted an operating point of 5 volts with a 9 volt supply, you'd use Ohm's Law to figure out what current would drop 4 volts over 10K, and design your CCS for that target current. And since E=I*R, then the current should be 4/10K, or 400µA. So, with a 400µA CCS at the source and a 10K drain resistor, the gain might vary, but the operating point will be forced to 5 volts regardless of minor variations in the JFET, and we don't have to select individual devices that will bias correctly or adjust a trimmer to get maximum headroom. I am also assuming from your schematic ( here comes the dangerous part) that bypassing the CCS with a cap will work just like bypassing a source resistor. Close, or still out chasing flyballs in the dark?

So, if that's all basically correct, all I have to do now is suss out the constant current source and figure out how it interacts with a bypass cap in regards to frequency response, and I can trade all those trimmers in for credit, right? 

[Transmogrifox]

Sounds like you've got it. The bypass capacitor and frequency equations aren't any different.  The CCS looks like an infinite impedance for any practical purposes, so all you need to consider in your RC time constant is the equivalent output resistance of the FET and the series resistor you are using.  Some JFETs have such low transconductance that the equivalent source resistance could be as much as 1k, though it seems more common to me that the typical would be a few hundred ohms.  If my memory serves me correctly, you can pretty well ballpark the output impedance as 1/gm.  I know it works for BJT's, and where my memory fails me is if the same type of reaoning works out for FETs.  I'm pretty sure it does, since the small-signal models don't differ much in that regard.

If you really want to get picky in your design, there are some optimums.  This is an optimal noise characteristic.

Thermal noise power on a resistor is sqrt(4kTRB), k=boltzemans constant, T=Temperature, R=resistance, B = Bandwidth.
This noise increases with increasing the value of the resistor.

Semiconductor noise for audio frequencies most generally problematic is 1/f noise.  There is also shot noise.  These noises characterristically increase with the current.

This is what sets up the balance: 
Real low resistance biased at a high current reduces resistive thermal noise, but increases semiconductor noise.
Large resistor biased at a low current is more heavy on thermal noise from the resistor and less on the semiconductor noise. 

It appears there's an optimum to be found between bias current and drain resistor.

In case you think it makes a difference in voltage gain, I did the math once.  For a fixed headroom, the drain resistor value does not effect the voltage gain much.  As the resistor gets smaller, bias current can be made larger, thus the device transconductance increases.  For a large resistor, bias current must be small, and the transconductance is much lower, making the end voltage gain about the same.  The only way to move up appreciably in voltage gain is to increase both the resistor and current, thus more headroom is required.

Another limit to the voltage gain is the load impedance.  If you're operating at cool bias currents, then output loading will make a significant difference in the gain.  It's a good rule of thumb to make sure that the total load current will not exceet more than 10% of the bias current.

If you keep the drain resistor in the  3k-10k range, then driving 500k loads is not a significant factor in the stage gain.

All that to say it doesn't require black magic to design with JFETs.  It's just that the CCS does consume a little more for parts...but then again It may be worth it to you to do away with the trimmers and be able to expect that your JFET circuit will bias up without any tweaking and tuning.  Headroom wil fluctuate with JFET types, but that's something you also deal with in the resistor-biased FET circuits.

[lovekraft0]

Thanks very much, it's all starting to make sense. I'll have to try this with a couple of other JFET projects. And thanks again for pointing out the FNJ598J, it's going to come in handy for those of us on a budget.

[Transmogrifox]

One word of caution with the FNJ598J:
Its absolute maximum drain current rating is 1mA (pretty low power device).  This will not be a problem for these 9V operated pedals with drain resistors around 10k. 

I kinda put this trem close to the ragged edge, but it's biased relatively low...and doesn't go near the ragged edge most of the time so I would say that it fits within the 80% safety margin. 

[Paul Perry (Frostwave)]

Since the original problem is lack of depth, would it be possible to simply subtract (that is, add out of phase) audio from the signal from the original EA circuit, cancelling the audio feeding thru in the 'low' part of the cycle?