There's a 2013 update on this project here: viewtopic.php?f=6&t=598
This thread is now locked. Please visit the 2013 thread.
There's a 2012 update on this project here: viewtopic.php?f=6&t=14&p=5980#p5980
No input capacitors or transformers in this one. Just straight wire to the IC input.
What I'm about to describe is a project I'm developing that is both on the drawing board and also on the bench. It is the result of out-of-the-box backward thinking. This is a work in progress.
While I'm out riding my bicycle I let the mind wander. So I thought about this: One of the problems with active Mic preamps, the almost universal need for input capacitors, has bothered me. We go to a lot of trouble to accommodate phantom power using two to four aluminum electrolytics or very large and expensive film caps. All this to make the microphone fit the DC input voltage confines of the preamp. Why? Put simply, because the preamp would blow up because it operates at lower voltages. So, as I'm sitting there on the bike trail thinking "why," I decide to think "why not."
How about making the preamplifier accommodate the microphone? I decided to take on the challenge of making a THAT1510 (or 1512) operate direct-coupled with DC inputs that range from 0 to +48 volts and all points in-between. This allows the input coupling caps, with their large stored charge, or transformers, to be eliminated.
Once this decision is made, with up to 48 volts DC on the amplifier inputs, it becomes obvious that nothing on the THAT1510 can be ground-referred. The power supplies and all connections to the preamplifer IC have to be "flown," or tracked to points that put the THAT1510's inputs within it's common mode range.
I immediately remembered a useful circuit I saw and prototyped in my youth.
See: http://www.ka-electronics.com/DC_Preamp ... 261974.jpg
This circuit, which I modified by making it a voltage follower to eliminate an input ground reference requirement, and resulting common mode limitation, was made to swing 100V P-P.
Those of you familiar with the JH-Series consoles will also recognize this circuit as the "Swinging Op Amp." I still remember that REP ad with the metal "2003" op amp sitting in a child's swingset. IIRC MCI used HA-2645 HV op amps and needed to replace them with 5534s. MCI developed this to make a 5534 run on elevated rails. Note that Q1 is drawn with the wrong polarity. It should be an NPN.
Image courtesy of MCI and http://www.proharmonic.com
The above circuits are pretty useful building blocks. The "swinging op amp" is a block we need for our preamp but it's only used as a DC amplifier.
When we connect a phantom-powered microphone to a mic preamp we usually have no idea what DC potential, reference to ground, will develop. As long as the mic is sourced enough current we don't bother to meter it to see what DC voltage develops. Transformers and coupling caps block that potential.
A direct-coupled preamp topology shouldn't care what the operating potential of the microphone is. But it's presence is significant. To determine the voltage at the input and adjust the preamp to match the microphone and it's quiescent operating potential, we integrate and measure the common mode DC voltage at the preamp input. This voltage is determined by the phantom source resistors (6K81 to 48 volts) and the microphone's DC current which produce an I*R drop. Thus, under operating conditions with a typical microphone connected, one might see voltages ranging from 12-44 volts at the input. With phantom off, this voltage is at ground potential. This quiescent voltage, we'll call it "Vref" is used to set the center-point "pseudo-ground" and supply rails of the amplifier and associated circuitry. The supply rails, typically 15 bipolar volts, are at Vref + 15 volts and Vref - 15 volts. Thus, the power supply rails and Vref pseudo ground voltages are set by the microphone and "slide" the common mode range of the preamp IC to fit the particular microphone and it's operating voltage. We call the tracking supplies "flying rails" because it sounds really cool.
Here's a schematic:
Update July, 11 2012: For updated schematics look here:
and here: viewtopic.php?f=6&t=14&start=143
1) We eliminate the input coupling caps and move them to the preamp output where they can be low-value high-quality film units. The LF cutoff of the preamp can be made very low, e.g. <1 Hz. This eliminates a source of LF CMR error and also prevents damaging stored charge and high peaks currents during faults.
2) We develop an input servo, to measure the microphone's Q-point, Vref, and use this servo in a feed-forward configuration (not the usual feedback servo topology) to establish a psuedo ground. This "ground" connects to the 1510s bias resistors and output Ref connection. We call it Vref.
3) We fit the DC input common mode range of the preamp IC to the microphone, not the other way 'round. We do this by flying the supply voltages up and down relative to the DC Q-point of the input.
4) We use a bootstrapped servo, to slew the Vref (psuedo ground) from "real" ground all the way to an insanely high +48 volts. IC supply voltages are +15 and -15 volts above and below this reference. The upper supply tracks up to about +65 volts.
5) We take the DC servo's bootstrapped "flying rail" supply connections to the THAT1510s supply pins so it also "rides the rails." This maintains our input common mode range and supply voltage requirements of the 1510 and avoids destruction. Other op amp stages to the left of the output coupling cap, the servos, are also powered from these rails.
6) We eliminate the large-value (2200 to 6800 uF) coupling cap in the 1510s Rgain line to improve LF performance and reduce capacitor-induced distortion. This provides an amplifier with full gain to DC where the microphone will most likely be the limiting factor in low frequency performance.
7) Because DC gain exists, offsets which occur within the microphone itself become significant. In many cases the input DC differential offset may be larger than the signal itself. Later in the post we explore the sources of microphone-induced offsets and deal with compensating it using servos. Without servos, the gain multiplication of the offset would cause the 1510 to clip. The servo samples input offset at the 1510s input transistor emitters (also known as Rgain) and correction is applied back to the input as an injected current. Further examination reveals that the phantom resistors and microphone internal resistors form a Wien bridge.
The preamp requires -18V, +18V, +48V, and +72 volt supplies. The highest supply must be at least 18 to 20 volts greater than 48V in order to provide a positive supply for the ICs.
Here's how it works: With a microphone common mode voltage of zero volts, dynamic or ribbon with phantom off for example, the THAT 1510's reference is ground; its supply voltages are roughly +/-15V and it operates like normal circuit. The 6K81 phantom resistors serve as Rbias.
With the microphone unplugged and phantom switched on, our inputs float up to +48 volts, the servo tracks it, and sets the THAT1510's reference to 48V. The 1510s positive supply voltage, Vcc, flys up to +63 volts, while Vee, roughly 30 volts below Vcc rises to +33 volts. A whopping +63 volts is applied to the positive pin of the IC, but it only sees 30 volt supply differential. The 1510s inputs, reference and output are at +48V. It thinks 48 volts is ground! Now connect a phantom-powered mic. With a realistic current drain from the microphone we may see an operating voltage of 12-45 volts. The output voltage of the 1510 with a mic connected: Same as Vref 12-45 volts. We then remove DC at the output with a good capacitor. There are no input capacitors to degrade low frequency CMRR.
Yes, this works and I've done it on the bench. The 1510 is still alive and amplifying very well. Tip and ring, through 6K81 Rs was checked from 0V to 48V with no problems. Inputs were repeatedly shorted to ground and each other to simulate cable faults without destruction. And unlike capacitor-coupled inputs there is no stored charge. The 1510 and servo op amps operate with a constant 30 volts from Vcc to Vee. The 1510 is able to drive about +18.5 dBv at all levels of DC input from 0-48V. Operating current, Iq+, ranged from 19.4 mA (48V out) to 23.6 mA (0V out). The MJE243, with a heatsink, has about 1W burned off at Vin=0. The MJE253 is also heatsinked.
Now there's one additional "sticky-wicket:" DC offset in the microphone, even millivolts, are amplified by the DC gain of the preamp. With 60 dB of AC or DC gain 1 mV offset becomes 1V. Typical offsets may be many times higher. Our fun however is not ruined.
Look at some microphone output topologies: Transformer, AC (capacitor-coupled) or maybe even direct-coupled. If a transformer is tied between the THAT1510's inputs, there will be no offset in the microphone nor any in the 1510. The DC resistance between pins 2 and 3 is low. But consider an AC internally-coupled microphone that pulls an exact 1 mA each from pins 2 and 3. This microphone is sourced to phantom through two 6K81 0.1% resistors. One resistor is exactly the right value, the other is 0.1% high. The offset voltage is 6.8 mV. With 60 dB of DC gain our output is now at 6.8V relative to "ground." (Wherever ground actually is.) No fun.
And then there's the obvious "offset" problem with shorted or mis-wired mic cables that produce a 48V differential. One would think pin 2 (or 3) shorted to ground would be a problem for the 1510. But examination reveals peak currents to be many orders of magnitudes less than a conventional preamp with stored charge on it's coupling caps.
No one said this would be a "cheap" preamp. It will have straight wire into the IC (with the usual protection circuitry), use a high precision gain switch, require +70, +48 and -20 volt supplies, and employee a high-quality film coupling cap on the output of the 1510 with a very low frequency cutoff of <1 Hz. A +/- 6dB variable level trim will allow precise adjustment between switch gain steps.
Installment Two: In this section I address issues relating to phantom power faults.
I'm thinking real world harsh reality. With the test circuit I found it to be very tolerant of patching and unpatching. As a test I ran it without protection diodes around the 1510. (Noise was consistent both before and after this torture test.) I could switch phantom on/off or short the input to ground with phantom applied. And, curiously, doing it while monitoring the output. Not the usual "pop" you get with capacitor-coupled inputs with lots of stored charge - just a little "whump." The servo is pretty fast, but when the input is shorted there is a time period for it to follow and equalize to the input.
I haven't calculated anything but I suspect that the actual energy dumped into the 1510 is less than what a capacitor would produce. The Hebert and Thomas AES paper, "The 48V Phantom Menace," shows peak currents of 2-3 Amps when simulating patching using conventional capacitor inputs.
This design benefits in that there's no stored charge.
I think I shall measure it with this topology - doesn't seem like it could be much worse than 2-3 Amps you get with 47uF capacitors under fault conditions. I'll fire up the storage scope do a differential measurement across the 4R7 and see if I can quantify it.
Later I decided to look at this another way: Assume a worst-case scenario where the servo is broken and stuck at 0 volts. The input to the 1510 is pulled down to this 0V through a 2K. Phantom is "accidentally" switched on and 48V is sourced through 6K81 to the 2K. This forms a voltage divider. Only 10.9V is developed at the 1510 input well within it's common mode range. Seems like very little current would flow at all.
Installment Three: I clarify to some forum members the role of the floating supply:
The floating supply is a DC amplifier. It only tracks the static voltage established by the microphone, the Q-point. The flying rail generator's job is to slide the rails up and down in response to the phantom potential not audio.
Example: With phantom off, Vq=0, the rails are at are ~ +/-15V (in the proto +/-13.4 with a 15 V zeners). With phantom on, and no mic to load the 6K81s, Vq is ~48V and the floating supplies are at +63 and +33 volts. With a microphone loading the 6K81s, say Vq=30V, the rails are at +45V and +15V.
Very important point. This is not like the MCI swinging op amp that was used to extend the output range of a 5534 to make it HA-2645-like. This allows us to use a low-voltage design and parts (except for the 100V transistors and 70V PSU). It allows us to extend the "headroom" of the DC parameters but not signal.
Installment Four: This is in response to one poster's very reasonable question "Why not just use a good transformer?"
As to being of any practical benefit in practice that remains to be seen. But, we spend a lot of time and energy eliminating electrolytic capacitors in other parts of the signal path and in this particular application, active mic input, is one area where I haven't seen input caps eliminated. Don't know why that should be off-limits. (I'm not saying there aren't any - just haven't seen one.) Should we be adverse to trying something new? If so, use a good transformer. Our grandparents did. I love transformers - you can't beat them for galvanic isolation and "free" gain among other reasons. But, it will cost more than the BOM of the entire preamp and it's already been done ad nauseum.
For active mic preamps the elimination of coupling caps has significant advantages not the least of which is LF CMRR (also a transformer attribute), leakage current, stored charge and perhaps, yet to be seen, transparency.
1) No complication in the rail generator - far less circuitry and about the same real estate than say, for instance, a JE-990. (No direct comparison functionally.) I count nine components with maybe a $5 BOM.
2) The AC performance is the same as a 1510 which is a reasonably good, if not an excellent monolithic performer. Transient response? It has a 15 MHz unity gain BW and close to 3 MHz at 60 dB gain. Slew rate: Same as the 1510 at 19V/uS typical. Ein at 60 dB gain should be close to 1 nV/sqrtHz. Noise figure? 1.3 - 1.6 dB with 200-150 ohm source R. Transient response? The 1510 is awesome in that regard. I had some 'scope photos but can't find them at the moment. Degradation from flying rails: None that I've been able to measure as it just sits there at Vq and Vq +/-15V. Improvement from elimination of input caps: Quite likely a lot. Certainly worth listening to with better mics than what I have.
What is it about a preamp that has 6-60 dB of gain range that makes it inflexible? Headroom? About the same as any preamp that runs off of +/-15-18 volt rails (could be +/-20 with a +75V supply) which represents virtually every modern console. If you don't want use 1510s then use something else. Virtually any low voltage preamp could be adapted discrete or monolithic. We shouldn't get too concerned or wrapped around the axle about audio degradation because the flying rail generator has no real effect on audio performance as it is a DC amplifer and in application never moves far, or fast, from the Vq of the microphone. And yes it works with phantom off too. In that situation it's just a 1510. (Note that the noise contribution of the servo, small, appears in common mode to the 1510's input. And, at the 1510's output Vref pin no more than a 1510 with any servo.)
EDIT: 9/19/06 Still some work needed on the input offset. I'm looking at using the Input DC servo THAT proposed in the Hebert and Thomas Paper "The 48V Phantom Menace." Willl keep all of you posted.
I think the offset problem has been licked as they appear to equalize in the last circuit I posted. Picking the servo input off of both inputs combined with lower value bias resistors THAT recommends for the 1510 allows the inputs to equalize. The upper and lower arms of the bridge formed by the 6K81 pull-ups and mic phantom take-off resistors are now bridged with 1Ks pulled to Vq. I couldn't find a way to get enough offset in it to matter as it just servo'd to where it needed to be.
Installment Five: There was a comment about transient distortion and transient response which I mis-understood. It resulted in a comparison between the SSM2017 and SSM2019.
The 2017 was a bit of a disappointment particularly after the 2016 which everyone seem to love. No nothing about the 2019.
For the record I do make THAT 1510s available and have sold many to replace SSM2017s. One comment I got from an engineer after a swap was that "the 1510 sounds so much brighter!" I asked brighter good or brighter bad. The response: "Brighter good - need more parts." Another response ran along the lines of "It's like someone opened the curtains that were in front of the monitors."
I did mis-read your transient distortion as transient response. The 1510 with it's dielectrically-isolated process is faster (and perhaps lower transient distortion) than even a 2019. Here's the chart:
Installment Six: I checked in with a progress report:
I made the flying rail generator op amp an OPA27. I measure +/- 5 mA output current from this IC so I'm not sure an LTC1012 is up to the task here. Much better choice than my OPA604 which was a quick option.
I have decided to make the +70V supply +72V which 24V times 3. More on that in a minute. The 70V supply current, Icc is about 18-20 mA. With Vz at 15V the negative supply can be lowered to -18V without any reduction in headroom.
I experimented with Vz being 18 volts. This puts 34.7V across the 1510 and OP27s. You get slightly more headroom but probably not enough to justify running the FRG OP27A with the added dissipation. (BTW the same if it was a line driver on regular supplies.)
I'm working on the power supplies now. The 48V and 70V aren't going to be that hard to get. I don't want to use doublers/triplers or AC-coupled bridges so I'm looking at two standard power transformers. One either 20 or 24 volt split secondaries and the second one 24 volt split secondaries.
I think I want +/-18V for the output buffers, +48 and +72. The two high-voltage supplies are very low current.
My goal is to not use any hard-to-find high voltage components e.g. TL783s or LM317HVs. Obviously the MJE243 and MJE253 are high voltage but relatively easy to get and fairly non-critical.
Installment Seven: The Power Supply.
Power Supply Updated 11-16-2012:
I've been working on the power supply for the DC Coupled preamp. My goal was to use readily-obtainable components to generate both the 48V and 72V rails without resorting special transformers, voltage doubler/tripler circuits, or AC-coupled bridges. For some reason, perhaps ripple current or excessive DA, capacitors seem to have premature failures when used as high-current carrying series elements.
Although the TL783 is a fantastic device for phantom power it is sole-sourced and sometimes difficult to obtain. Since the flying rail generators in the preamplifier itself are a form of DC bootstrapping, I decided to continue that tradition and bootstrap 24 volt supplies to provide both 48 and 72 volts. This allows the use of low-voltage regulators and filter caps.
The output buffers, not yet finalized, are to be powered from +/- 18V. The negative supply for the flying rail generator is also powered from -18V.
My goal is a power supply for a dual preamp unit. Approx 50 mA at 72V and 25 mA at 48V are required.
I used the LM317 and LM337 because they are superior to the 78XX/79XX series and permit up to 40 volt input/output differential. With the 24V secondaries used under high-line (130V) unloaded conditions, the input to the regulators approaches 40V. This is the absolute maximum for the 7824 and above the 35V maximum of the 7818. I also have confirmed good low-line operation below 100V.
Here's the schematic:
You can see the three 24V supplies stacked on top of one another to yield 72V. An added bonus is that the 72V supplies' dissipation is spread among three regulators, the 48V among two. The 10K pre-load resistors don't seem to be necessary but I put them in for good measure.
Now that I'm no longer running on bench supplies I'll move on to the output buffers. More to follow.
Installment Eight: This post was in response to "Why not use good coupling caps?"
I did look at that originally and the biggest poly [anything] I could find was around 50 uF. There might be bigger ones, but I looked at Electrocube. See: http://www.electrocube.com/pdf/cap7.pdf
So yes it could be done with input film units and IIRC Roger Foote used big films on his Pico Preamp. There's an awesome picture of his build See: http://www.prodigy-pro.com/forum/viewto ... p&start=19
I think that film input caps (Two or four 50 uF in 1%) in DIY qty might be as much as a tranny.
I wanted to eliminate input capacitors for a number of reasons beyond the obvious sonic limitations electrolytics might impose. Some of these reasons are also shared by polyfilms: Two of them are stored charge and CMRR.
Stored charge is really hard on input devices and protection diodes. THAT measured 2-3A peak both into inadvertently-connected line outputs and device input protection diodes. (And this can damage low-noise performance.) This preamp has maybe 10-20 mA peak current into the input and tolerates tip/ring shorts to ground with ease. You can't draw sparks with it. In a conventional preamp, the bigger the input Cs, the bigger the spark. There's something about connecting an arc welder to a delicate microphone and preamp that's always annoyed me.
CMRR: Big polypros (both value and size) with reasonably low input Rs on the 1510 are still going to be hard-pressed to have much better than a 10-20% value match at reasonable cost. Electrocube do have 1% but I was afraid to price them. A preamp could have a LF CMRR trim but tracking still wouldn't be great.
Besides, capacitors have been used before yes? I wanted to do something different. Something electrically more like a transformer whose tip and ring primary connections can float up and down.
What does a mic preamp with subsonic response sound like on kick drum? (And yes a HPF will be needed on other instruments. I'm sure that "P-pops" will require one.) And, at least in the input, zero LF phase shift. I do use a film at the output with a very low corner frequency so the phase shift there should be quite small.
As to parts count the BOM cost of all the semis is a fraction of the cost of a good transformer. The gain stage has three op amps, two of which are DC devices. The input Vos servo, by virtue of it driving Vactrols, doesn't have to be low noise anymore - just DC precise. And we don't have to use a 1510 or even a monolithic device for the gain stage. It could be discrete for uber high-end.
Installment Nine: This was some prior art that was sent to me regarding flying rails:
Thanks to everyone who sent me the JAES Feedforward Floating Power Supply cite. I don't feel right about linking directly to the pdf on this one but I think figure 7, below, is not a significant "taking" of the work. If anyone objects to it being here I pull it down. But for now:
And if I'm feeling really bad and receive enough encouragement from the people who sent it to me I'll host the whole thing. Is it available online? Would I be diminishing the value of the original?
Now isn't this interesting. This circuit provides some wonderful slew rate improvements to a common op amp. They made a 4558, yes a 4558, slew at 300 v/uS. We could still use this trick. And perhaps we do...
Installment Ten: I discuss Mic Output Topologies. The concern is due to DC offset that a microphone may internally generate.
Mic output topologies
AC-coupled topology, high DCR between pins 2 and 3. Resistive imbalance possible.
Schoeps topology below. Direct-coupled self-biased emitter follower. Resistor tolerance, zener and transistor matching errors. The emitter resistors for Q2 and Q3, not shown, are the 6K81 pullups located in the preamp.
Aside: Do you suppose we should view at the Schoeps topology as a "current loop" interface? It looks suspiciously like an industrial 4-20 mA current loop when you think about how the emitter load, the 6K81 phantom pullup, is at the far end of 100 feet of mic cable sourcing current into the mic which is sinking a variable amount of current. The output from the microphone is a current converted to voltage by the 6K81 at the preamp input. Didn't see this before.
Neumann U89. Uses balancing resistors to create a "phantom" transformer center tap. Some resistive error but bridged with a very low DCR transformer.
Sony C38. Uses a real transformer center tap. Winding DC balance but bridged by such a low DCR that it probably doesn't matter.
These are representative samples of what I've found so far. There may be even more topologies.
Of all of them the direct-coupled emitter follower is probably going to be the most problematic.
Installment Eleven: Some pictures of a mess.
It was warm in the shop today. So I took some pictures. In a thread I've made some comments about breadboarding. It's dangerous ugly, untidy work. But somebody's got to do it.
Installment Twelve: I had gotten sidetracked using LDRs as servo correction devices. I went back to a voltage offset servo.
I did try Vactrols to DC-balance the bridge and correct offset. They worked great and I placed them in the bridge arm so that if a DC imbalance was created by a resistive imbalance in the mic, the error was partially corrected by the Vactrol. It worked very well. I think everyone agreed that voltage offset, actually current injection, was a preferred method. Given that, I re-explored the servo op amp looking for a better DC performer with low bias current. It was suggested that I try an OP07 in addition to an LT1012. So I did. OP07's are cheap too.
I've just completed evaluating different servo op amp choices replacing both OPA27s (because that's what I had) with OP07s and LT1012s.
I modified the circuit to remove the LDRs for offset correction and return to resistive current injection using a 10K from the servo output to the 1510 inverting input. A 10K from the non-inverting 1510 input connects to Vref derived from the Flying Rail Generator. The second restores CMR balance by way of equal resistive loading. The schematic is here:
Using resistive injection permitted easier measurement of what the servo was doing because it's output did not have to overcome the LED threshold.
The first step was to check the Flying Rail generator with an OP07. The integrating resistors which sample tip and ring voltages are large: 499K. This provided a fair amount of offset with the OPA27 due to it's bias current. The OP07 improved that considerably and those resistors could easily be lowered in value. I also checked the output voltage compliance of the Flying Rail Generator and it measured the same as the OPA27 does.
To simulate an input offset I used 6K81 1% phantom pull up resistors with 10K 5% pull downs to ground. This produced an approximate 60 mV input offset (uncorrected) with the random resistors I chose.
With an OP07 as an input servo, with 100K servo resistors as shown, the corrected input offset measured at the 1510 input was 280 uV.* The 1510 gain was 60 dB. (*Stressing the limits of my Fluke 8050.)
Looking at it another way, I measured the difference between the 1510s Vref and Vout. With input servo correction, I measured a difference, an "output offset," of ~-60mV at 60 dB gain. Without correction applied the output saturated to the positive rail.
I also tried the LT1012 in the input servo and the sample I had produced slightly higher input Vos. But still good enough.
Yes, the OP07 makes a fine set of servos at $0.70 each. Much better than the $5-ish LT1012.
I could not hear any noise floor difference between a servo'd preamp and an un-servo'd one.
And with the OP07 input servo there were very, very little switch clicks.
So I guess it's time to re-draw and finish the output stage/gain trim/mute/polarity stuff.
Installment Thirteen: Some Housekeeping.
Here's an updated schematic:
Still kinda messy but I'm getting the drawing layout dialed in. A redrawn Flying Rail Generator may make things a little more clear.
The latest uses OP07s for both the input servo and Flying Rail Generator. I've lowered the resistor value at the FRG input and scaled the cap up to maintain a ~50 mS time constant. Note the 10K resistor on the 1510 from pin 3 to pin 5. This establishes CMR balance to compensate for the 10K servo injection resistor at pin 2. It's quite likely that the values of these can be raised considerably once microphone offset characteristics are better known.
A quick summary for those who are just now joining us:
The flying rail generator has three outputs: +VFly, -VFly and FlyRef. Flyref is at the output of the top OP07 and drives the 1510's Vref, the CMR compensation resistor and the 0.1 uF integrator cap on the OP07 input servo. With phantom switched off FlyRef is at ground. With phantom on and no microphone connected it rises to ~48V. With a microphone connected, FlyRef rises to whatever voltage the microphone sits at. +VFly and -VFly are approximately +/-15V relative to FlyRef. These supplies feed all active circuitry to the left of the output coupling cap. The operation of the Flying Rail Generator is key to the elimination of input coupling caps. It's the heart of the circuit.
As soon as I can get a better drawing up I will.
Installment Fourteen: More Housekeeping.
I've run a few more servo tests and performed a minor tweak.
The first thing I did was raise the Flying Rail Generator zener voltages to 1N4745 16 volts. This provides a 30V tracking supply for the 1510 and OP07s and increases 1510 headroom.
I'm still debating what values to use for the "low voltage" bipolar supply. Either +/-18V or +/-20V. The -18V is a little tight for the Flying Rail Generator with 16V zeners; -20V may be a little "hot" for the op amps to the right of the coupling cap. Still thinkin' on that. It's a headroom/op amp heat issue.
I decided to look at how much correction current the OP07 provides to the 1510. Using my randomly chosen 10K resistors that produced an ~60 mV uncorrected input offset error (2K Rterm) I measured the output current of the OP07 with various values of Rterm. The current was measured indirectly by reading DC volts across the 10K.
The 10K pull-down resistors with 2K Rterm (60mV) required 71 uA correction current.
With a 150 ohm Rterm the uncorrected Vos was ~6 mV and 135 uA of correction current was required.
As a reality check I reversed the resistors and observed a sign change in Ios. No great suprise but imagine the fun I would have had if there wasn't one.
With the 10K puldowns removed, and essentially no deliberate Input Vos occuring, I measured a 2.7 uA correction current regardless of 1510 gain.
Now if 60 mV of offset is a realistic microphone-induced error, and that's a very big "if," then the 10Ks could be made much larger. There's still plenty of compliance range in the OP07 servo. If 60 mV is a small real-world error, then with 10Ks there's still plenty of correction available. Point is we're only using about 1.35 volts (max in my tests) swing from the OP07 output. There's at least another 9-10 volts left in each polarity.
So the question remains. How much is enough? My thinking was the Direct-Coupled PNP Emitter follower topology would be the worst offender.
Installment Fifteen: Squarewaves!
I had been looking for these 'scope photos of the THAT 1510 for months now. I probably didn't find them as soon as I could have because they were in an SSM2016 folder. Doh!
When I first got samples of the 1510 I wanted to check the squarewave response at high gains. This is where the SSM2017 really began to crap out on bandwidth. So I did a 20 KHz squarewave test at both 40 and 60 dB gain as well as a 200 KHz torture test to see how well the part behaved. Now I didn't expect 200 KHz at 60 dB gain to be pretty. In fact I thought it might come unglued. But it didn't. The 1510 is very well behaved. Having a 3 MHz -3 dB bandwidth at 60 dB gain pays off.
Installment Sixteen: Input RFI Filters.
We discuss input RFI filtering focusing on common mode chokes and rolling off response at the input rather than letting it pass through the gain stage. The discussion began with the relatively high cutoff of the input RFI filters in many designs. I omitted the common mode choke discussion because it had so many dead ends with US vendors.
I don't think the 2019 application note would make you any happier:
(Thanks for the drawing Sam Groener.)
We also discuss common-mode chokes and I basically came to the conclusion that there are a few good choices but none stocked by distribution or actually available in small runs. This explains why BenchMark have their own CM-1 made.
Installment Seventeen:The Output Stage. Everything to the right of the single coupling cap.
OK, I finally got around to checking the output stage on the Protoboard:
Updated 11-16-2012 here: viewtopic.php?f=6&t=14&p=6310
I wanted to keep everything on the right side of the coupling cap as simple as possible. But I wanted it to have the following features:
High impedance buffering of the coupling cap
Switchable low cut
The 1 uf NP film I've chosen for my prototype is a Vishay/Roederstein polypropylene and I wanted it lightly loaded. With no LF cut, the corner is about 2 Hz. I suppose that for low squarewave tilt we could use a 1M bias resistor with a noise penalty. Since the bias resistor is rather large, a bifet would appear to be a good choice and I've chosen the OPA2604. It could be just about any decent dual Bifet. The other two corner frequencies are approximately 20 and 80 Hz.
The second stage provides a +/- 6dB gain trim which, in combination with the 1510's stepped gain switch, allows fine gain adjustment to span switch steps which may be as high as 5 dB/step. As many of you know I am not a big fan of op amp bias current being allowed to return exclusively through a pot wiper. As the pot ages, momentary lifting of the wiper during rotation could create a brief open-loop condition. The 220Ks from ends to wiper prevent this. If the pot completely opens, the stage goes to unity gain. These resistors could also be modified to "bow" the pot's taper or provide an accurate center detent.
I chose a THAT1646 for the output buffer and polarity reversal is provided at the 1646's output. The bridge rectifier provides output protection in the event that the preamp is accidentally patched into a 48V-powered source. The bridge simply replaces the four output diodes extending from the outputs to the rails as is normally seen in the THAT and SSM application notes. An unbalancing switch is provided for single-ended connection. If low Common Mode output DC offset is a concern, two 10 uF BP caps may be placed in the 1646's sense lines. These caps are not directly in the signal path and are lightly-loaded by the 1646. The H, or + output is taken from the 1646's inverting output to re-invert the second stage. Thus absolute polarity is maintained from 1510 input to 1646 output.
With the second stage at maximum gain, the combined gain is +12 dB: 6 dB is provided by the THAT1646. Thus, the total preamp gain will be +12 up to +72 dB.
I would prefer to provide polarity switching in the second stage and am exploring that possibility. My desire is to keep the signal path short but maintain great LF response. Thus, there are two intermediate stages: One to buffer the cap, the second to provide fine adjustment of gain with linear control.
EDIT: There are rail clamps at pin 3 of the OPA2604: Forgot to draw them.
Installment Eighteen: More about offsets.
I'm curious to hear what you find out about real-world microphone differential voltage offsets. I've heard rumor of bizarre asymmetrical output topologies and wonder how they would affect your circuit, but I don't have any direct experience with them. I hate to name names because I'm going off vague memories of anonymous old usenet posts, but I want to say that the TLM103 maybe has an asymmetrical (AC) output and that the Behringer measurement mike (which people seem to like on drums) has an intentionally asymmetrical phantom current draw. Don't quote me on this, but I might suggest these two microphones for a closer look.
I did correspond with Neumann about this recently and they pointed out that although the IEC specs only the tolerance of the phantom sourcing resistors, there is no spec for the microphone's internal Rs at this time. They cited <0.4% mismatch defined in the IEC 61938 and went on to say that most manufacturers also adhere to this tolerance on the mic-side.
In a later e-mail they mention: "Yes, there one or two manufactures sending all the current over one pin. All the others are more or less symmetric regarding current."
From what they said I don't think the TLM-103 is one of them. But, who does use single-pin feeding?
I think the Behringer measurement microphone, a favorite on drums I hear, is one of them.
This might be one situation in which a Vactrol-based counterbalancing servo resistor is appropriate. In the prototype I should have a servo "mode" switch from current injection to delta-R Vos correction. My initial thoughts are that these microphones (the Behringer ECM8000) would probably have a CMR imbalance built-in due to their unequal resistive/current loading. Somewhere I've seen a post where the Behringer were actually identified as being unbalanced.
Installment Nineteen:I move on to LF squarewave measurements. Without input coupling capacitors and a <2 Hz LF cutoff it should be very good.
I've been promising everyone low frequency squarewave photos to demonstrate the low tilt.
This shot, very difficult to shoot and somewhat poor visual quality is 20 Hz at ~ 60 dB gain. The scale is 5V/Div.
The input Vos servo values are 100K/100nF. The coupling cap at the output of the 1510 is 1 uF loaded by 1M. The 'scope output is taken from the first OPA2604 output. Thus we are seeing the LF tilt performance of the input servo and coupling cap. All stages which follow this are DC-Coupled so this should be representative of its end-to-end performance.
You can see a little RF riding on it. (Everything is open lead.) The baseline offset you see is simply 'scope V position. The generator I'm using for this test is a DC-Coupled battery-powered fully-floating thingy I have from Chung Instruments. So the preamp is being fed from a floating "balanced" source.
By changing the Vos servo components the tilt could be reduced further but I'm not sure it's worth it.
EDIT: I just double-checked and part of the baseline shift is due to the generator duty-cycle not being quite 50%. But still very low tilt.
Installment Twenty:I look for something to use other than the OPA2604.
I'm looking for alternatives to the OPA2604 (or OPA604) to buffer the 1 uF coupling cap. Why? Because there may not be any OPA2604s made for awhile and it's always good to have a second source. Deliveries have now been slid to another 26 weeks with the delivery currently September.
Here's my current alternative, the AD823:
http://www.analog.com/UploadedFiles/Dat ... /AD823.pdf
With a 1M bias resistor following the 1 uF coupler our choices are pretty much limited to Bi-Fets. Ibias is significant from a DC perspective and Inoise from a noise perspective. I really like the idea of maintaining low tilt so I'm going with a 1M.
Consider a common alternative, the OP275. An OK part perhaps, but the Butler combo JFET/Bipolar input has a pretty high Ib. Use the 350 nA max (25C) figure with a 1M and you've got a Vos of 350 mV. Not good. Maybe OK with a 100K Rbias. So we can rule this part out from the get-go.
The OPA2604 has a pretty impressive Ib of 100 pA but that's only typical. No max is given. So figure a typical offset from Ib of 100 uV. In reality input Vos itself (+/- 5 mV) will appear to dominate. And from a noise perspective it's OK. But the OPA2604 is made from unobtanium.
Honestly I had never looked at the AD823 but it's ADI's suggested replacement for the OPA2604. And maybe it's a good one. First off it's available in through-hole. Although it doesn't have the output drive of an OPA2604 nor the 48V supply I don't need that. It looks to have good bandwidth, unity gain stability, OK voltage noise, good current noise and well-characterized DC with low Ibias.
I searched the forum and haven't seen any references to the AD823. Should I consider them to replace the OPA2604?
Installment Twenty One: A forum member makes the point that the buffer stage is a follower.
While I like the high-impedance input of this stage, the common mode issues with it being a follower do bother me from an input non-linearity/distortion viewpoint. So what to do. Put gain here or make it an inverter? I place my bets on keeping the coupling cap lightly loaded and finding a better follower. Additionally, availability of the OPA2604 is still a big question mark.
I've just ordered AD823s to test them. In the AD823 datasheet they talk about using input resistors to limit current in the event an input is taken higher than the supply rails. But there may be an AC justification for it as well: Walt Jung talks about balancing the source impedances to overcome non-linearity in voltage followers here:
http://www.analog.com/library/analogDia ... inal_I.pdf
(See pages 54-55.)
Also in the AD743 datasheet:
http://www.analog.com/UploadedFiles/Dat ... /AD743.pdf
(See page 11.)
So I have a couple of options if I want to improve the performance of this non-inverting stage: One add gain. Two, keep it unity but add a couple of resistors about 1K or so in series with each input. In the case of the AD823 they are required to prevent overdrive if the input exceeds the rail by 300 mV. This could happen during phantom on/off transients. (Although not shown I do expect to used diode input clamps.)
http://www.analog.com/UploadedFiles/Dat ... /AD823.pdf (See page 15.)
So I'll continue to mull this over while I wait for UPS to deliver my ADI parts.
Installment Twenty Two: I compare the ADI 823 to the OPA2604.
I've had a chance to try the Analog Devices AD823 as a substitute for the OPA2604 in the voltage follower stage. The '823 is a very nice part.
Both devices were compared as unity-gain followers with 1K "balancing" resistors in series with the inverting and non-inverting inputs. Tests were done using +/- 15V supplies; the actual circuit will use +/-18V rails. The input bias resistor was 1M.
Both devices' THD were similar: They were in the measurement mud at ~ -105 dB 2nd and 3rd. The dominate distortion contributed by the device, which was very low, and appeared to be third harmonic. Both added similar amounts to the generator residual which was an approx. 3-4 dB increase. Noise was also similar and it too was buried in the mud around -110 to -115 dB in the 10-20 KHz octave.
With 15 V bipolar supplies the OPA2604 delivers 25V P-P before clipping; the AD823 is essentially rail-to-rail at 30V P-P.
Unfortunately the AD823 exhibits polarity reversal in this configuration. Very slight amounts of positive clipping (overdrive) produce a negative-going reversal to the opposing rail.
The OPA2604 clips slightly asymmetrically at 25V P-P but does not exhibit significant polarity reversal. As the output advances beyond clipping there is a slight momentary attempt at reversal. There appears to be some internal correction which brings the '2604 out of reversal which also results in a slight base-line shift. Clipping however is somewhat rounded - almost tube-like.
If TI can actual get around to producing Burr-Brown parts again the OPA2604 will be my first choice given that it clips cleaner. If the recipe for the OPA2604 continues to be lost the AD823 is indeed an acceptable substitute. The only thing that the AD823 has going against it is polarity reversal.
With the work that I've put into eliminating AC-coupling at the preamp input I wanted to make sure that the following stages didn't limit performance. It's good to know that I don't have to depend on TI if we've seen the last of the OPA2604s.
Installment Twenty Three:More tests on the follower stage.
I've run some high level 100 KHz tests to see if I can sort out some additional differences between the OPA2604 and AD823. I chose such a high frequency since my measurement fundamental with RMAA is limited to ~6 KHz. As Douglas Self points out on his site, the OPA2604 has rising HF distortion. I wanted to see if it was even more pronounced at 100K where I could see it.
Since I'm working at 100 KHz my choice of equipment becomes a highly-modified Heath IG-18 as generator and HP-334A as analyzer. At this frequency the generator residual is about 0.06%. Not great but as it turns out adequate.
The power supply rails were chosen to be +/- 18V with a 10V P-P output. The stages are followers with 1K "balancing resistors" of Rin and Rfb. The output is loaded in a high impedance with a 100R build-out.
The OPA2604 had 0.25% THD+noise at 10V P-P and 100 KHz. It performed it's best at lower test frequencies with a 1K Rin as opposed to 0 ohms. It really didn't care much either way if Rfb was 1K or 0 ohms. The 100KHz THD was primarily 2nd harmonic. At 25V P-P and 100 KHz the OPA was much worse, approaching 0.5%.
Bear in mind that the OPA test sample was an original Burr-Brown part from the last century during the golden era. (1993 date code.)
The AD823 was at the generator residual so it's THD contribution was immeasurable and below 0.07% at 100 KHz 10V P-P. As to the balancing resistor values I could not see much difference at 10V P-P. At 30V P-P, there was a very slight increase over generator residual. At this level, the AD823 had a slight edge using a 1K Rfb. It really didn't care either way what Rin was.
The AD823, despite it's polarity reversal at clipping, performs far better than the OPA2604 in measured THD at extreme frequencies. Given the OPA's dominate 2nd harmonic component I'm not sure it will sound much worse at more realistic frequencies.
Now about the polarity reversal of the AD823: Given that the THAT 1510 preamp and servos are operating at +/-15V (not cast in stone but based on conservatism) its' clip point, at ~26V P-P will be lower than the AD823s 36V P-P clip point. The 823 will swing within 25 mV (unloaded) of the rails. In actuality it will be loaded in ~10K in the first stage and 5K (the THAT 1646 output buffer) in the second. So it will come darn near 36V P-P. Given that I'm not sure that it will ever be hit hard enough to actually clip.
I planned to add a clip indicator with hold anyway.
So the choice becomes do you want the cleanliness of the AD823 or the "tube-like" second harmonic and smooth overload (at ~30V P-P) of the OPA2604.
I have to say that I'm very impressed with the AD823. I don't know why people don't write about it much. They are expensive: $5 in singles at Digi-Key. And you can get them.
Installment Twenty Four:Some prior patent art.
I found a couple of interesting patents:
This one discusses floating rails and uses an all NPN implementation to fly a voltage follower. Seems to acknowledge the prior art. (see figure 2.)
http://www.ka-electronics.com/Images/pd ... Supply.pdf
This one is an application publication in which floating rails bracket the potential around an input. Very similar. The inventor resides in Limerick Ireland and patent counsel is in Waltham MA. Couldn't see any reference to ADI but I'm sure that there's a connection there. It appears to be instrumentation-oriented.
http://www.ka-electronics.com/Images/pd ... 087364.pdf
Installment Twenty Five:An update.
It's time for an update on the project. I redrew the output stage incorporating changes made since trying the AD823.
A larger drawing is here.
The first order of business was to add clamp diodes to protect the AD823/OPA2604. Although 1N4148s may be fine here I'm using 1N4004GP glass passivated diodes since they will also be used as 1510 protection.
An Alternate Means of Gain Control:
I was cleaning an old piece of gear I made and was looking at the thumbwheels on it. I thought: "Who would use thumbwheels anymore?" "This is ancient technology." Then I thought: "I would."
It hit me. Why not consider the ancient thumbwheel switch as a means of precisely setting mic preamp gain? These have gold plated contacts and have lasted 30 years in my Mom's kitchen timer. In the studio one could know exactly what gain, to the dB, was being used. Great for repeatability and pencil-based "total recall." Didn't the Sphere or Quad Eight do this on the EQ with lever switches? Since our preamp doesn't have up to 99 dB gain we might want to black out those positions above "6" on the first digit.
Pardon the gradeaux - I think it may be cake mix.
For studio use substitute pizza...
Now we've all seen some very fancy relay-based preamps or digitally-controlled shaft encoder versions. And there's also TI's PGAs. There's one I've seen with a shaft encoder and Blue LEDs. It's real pretty. But lots of hardware is required to vary a resistor.
Many of these designs are based on having 60 or more 1 dB steps and lots of relays. Why not consider an "instrumentation" approach with the left-hand switch setting the 1510/1512 gain in dB decades and having the right hand switch set the gain, in single dBs, in the follower. Sure sounds simple.
Although I suppose one could get creative and use BCD switches calculating the Rs "one of ten" switches are still available. Digi-Key have them for around $14-16 dollars each in Cherry.
There might be issues with capacitive coupling between sections. Perhaps a copper Faraday shield might be needed between them.
So now we have three gain adjustment options:
1) Single Rev Log Pot (jumpy)
2) 12 Pos Rotary switch with +/- 6dB post-pre trim. (OK and offers infinite fine adjustment.)
3) Precision thumbwheel switch. (Gain in precise 1 dB chunks with "recall.")
I think I will include this added build option in my gain calculation spreadsheet.
Does this make sense to anyone except me?
Installment Twenty Six: More Prior Art on Flying Rails sent by a forum member.
Image Courtesy of Linear Technology
Looks pretty familiar. I think my original EDN cite was 1974 and here we have the same thing 20 years later in '94.
As many of you know, the DC preamp's flying rails are not signal driven (they sense the DC common mode voltage and are driven by a slow DC servo) but for signal driven applications where the op-amp is slewing AC Mosfets would seem to be a very good choice due to their reduced drive current. IIRC the servo drive current requirements are around +/- 5mA. Even so, they would be fun to try both in the servo and in headroom-extension applications.
Thanks for digging this one up. Maybe with enough prior art we can keep B*ringher and their ilk from patenting this.