From: Fred Bloggs on


Jim Thompson wrote:
> On Wed, 28 Sep 2005 08:44:46 +1200, Terry Given <my_name(a)ieee.org>
> wrote:
>
> [snip]
>
>>Dostals method also allowed me to
>>directly calculate the phase margin.
>
> [snip]
>
>>Cheers
>>Terry
>
>
> What IS Dostal's method? The Loop Gain & Phase analyser on my website
> is based on R.D. Middlebrook's laboratory technique, and is VERY
> accurate, since the loop is never actually broken.
>
> ...Jim Thompson

Middlebrook is still going strong too:
http://ardem.com/index.asp

From: The Phantom on
On Thu, 22 Sep 2005 09:53:22 +0200, "Fred Bartoli"
<fred._canxxxel_this_bartoli(a)RemoveThatAlso_free.fr_AndThisToo> wrote:

>
>"The Phantom" <phantom(a)aol.com> a ?crit dans le message de
>news:7sr3j157pdfap23guks1go3p9g7qjrvn5k(a)4ax.com...
>> On Wed, 21 Sep 2005 17:36:57 -0500, John Fields
><jfields(a)austininstruments.com> wrote:
>>
>> >On Wed, 21 Sep 2005 21:03:48 +0100, John Woodgate
>> ><jmw(a)jmwa.demon.contraspam.yuk> wrote:
>> >
>> >>I read in sci.electronics.design that "RST Engineering (jw)"
>> >><jim(a)rstengineering.com> wrote (in <11j3cjlkfbcqj6e(a)corp.supernews.com>)
>> >>about 'Op Amp Calculations', on Wed, 21 Sep 2005:
>> >>
>> >>>However, I can't convince myself that I can mathematically come up with
>> >>>the resistor value. I have googled the problem and come up short.
>> >>>Anybody got a pointer to a URL that goes through the math of how this
>> >>>configuration works? And what I'm doing to my phase margin?
>> >>
>> >>I believe Win has sort-of proprietary rights on this circuit. But it's
>> >>easy if you re-draw it. Why do these things come up in s.e.d. instead of
>> >>a.b.s.e?
>> >>
>> >>ASCII art; use Courier font.
>> >>
>> >> _______R4_______
>> >> | - |
>> >>o----R1--+---|\ |
>> >> | \_____ out |
>> >> | / | |
>> >>o------------|/ R2 |
>> >> + |----'
>> >> R3
>> >> |
>> >>o--------------------+------ Gnd
>> >
>> >---
>> >
>> >I believe the circuit he described looks more like this:
>> >
>> >Vcc------------+
>> > |
>> > [R1]
>> > |
>> > +----------|+\
>> > | | >-----+-->Vout
>> >Vin>------[R3]-------+----|-/ |
>> > | | |
>> > [R2] +-[R4]-+-[R5]-+
>> > | |
>> >GND>-----------+-[C1]--[R6]-+ A
>> >
>> >
>> >Where, for DC,
>> >
>> >
>> > R4 + R5 Vcc R2
>> > Vout = -Vin --------- + ---------
>> > R3 R1 + R2
>> >
>> >For AC, however, (assuming Vin and Vout are perfect voltage sources)
>> >we wind up with Vout trying to force the - input of the opamp to
>> >stay at whatever voltage the + input is set to by R1 and R2.
>>
>> For AC, assuming that C1 is very large so that it can be treated as an
>AC short, and
>> the op amp gain is infinite, the gain is:
>>
>> -((r5*r6 + r4*(r5 + r6))/(r3*r6))
>>
>> If A is the op amp gain and C1 is included in the calculation, the gain
>is :
>>
>> -A*(r4 + r5) - A*c1*(r5*r6 + r4*(r5 + r6))*s
>> -----------------------------------------------------------------------
>--
>> r3 + A*r3 + r4 + r5 + c1*(r5*r6 + r4*(r5 + r6) + r3*(r5 + r6 +
>A*r6))*s
>>
>
>Well, you've forgotten the GBW in all that :-)

No, I didn't forget at all; it's implicit in A.

>If we set WT=2.pi.GBW then we have
>
> -A - B p
>----------------- with
> 1 + C p + D p^2
>
> A0(R4+R5)
>A = ----------------
> R3(1+A0)+R4+R5
>
> A0.C1(R4 R5 + (R4 + R5) R6)
>B = ----------------------------
> R3(1+A0)+R4+R5
>
> (A0(R3 + R4 + R5 ) + C1.WT((R3 + R4)R5 + R6(R3 + A0.R3 + R4 + R5))
>C = --------------------------------------------------------------------
> (R3 (1 + A0) + R4 + R5 ) WT
>
>
> A0.C1 ((R3 + R4) R5 + (R3 + R4 + R5) R6)
>D = ------------------------------------------
> (R3 + A0 R3 + R4 + R5) WT
>
>
>Less than 2 min to work this out from scratch, incl. sign error correction.

What sign error is that?

>Isn't that Mathematica lovely? ;-)

Unfortunately, the result doesn't seem to be correct. Put the
expressions for A, B, C (the numerator of the "C" expression seems to
be missing a closing parenthesis), and D into the first expression,
namely:

-A - B p
-----------------
1 + C p + D p^2

Then when you have it written out in all it's glory, find the limit
as A0 --> infinity, and then the limit as C1 --> infinity. You
*should* get: -((r5*r6 + r4*(r5 + r6))/(r3*r6))

But, alas, you don't.


From: The Phantom on
On Mon, 26 Sep 2005 10:42:52 -0700, The Phantom <phantom(a)aol.com> wrote:

>On Wed, 21 Sep 2005 12:20:11 -0700, "RST Engineering \(jw\)" <jim(a)rstengineering.com>
>wrote:
>
>>A lot of years ago (38 to be exact) a wise old "rule of thumb" engineer
>>taught me a slick way of getting maximum gain out of an opamp without
>>resorting to very high or very low values of resistors.
>>
>>As we all know, for an inverting opamp, the gain is given simply by Rf/Ri,
>>where Rf is the feedback resistor from output to inverting input and Ri is
>>the resistor between signal and inverting input. The DC level of the output
>>may be set anywhere you choose by an appropriate bias level on the
>>noninverting input. For AC amplifiers from a single supply, this is
>>generally Vcc/2 with capacitive coupling between Ri and the signal.
>>
>>However, for very large AC gains, either Rf must be rather large or Ri
>>rather small. Rf being rather large makes the input voltage/current errors
>>become significant as regards quiescent DC output point and Ri being rather
>>small requires large capacitors for coupling and loading errors from the
>>signal source.
>>
>>So, sez old wily rule-of-thumb, just break Rf into two reasonable sized
>>equal value resistors equal to Rf/2 and run them in series from output back
>>to (-) input. And, from the midpoint tap on these two resistors run a
>>series RC circuit to ground. Bingo, the AC gain improves greatly.
>>
>>And guess what, it works. How do I calculate the R in the series RC circuit
>>I asks old wily. The answer comes back "Tweak it until you get the gain you
>>want." (Assume that C can be made appropriately large to get the
>>low-frequency gain you want.)
>>
>>I haven't used that trick in an awfully long time, but I've got an
>>application that needs it. And, if I want to use Diddle's constant in a
>>simulation program I can fool around (ahem, heuristically experiment) to get
>>the gain I need.
>>
>>However, I can't convince myself that I can mathematically come up with the
>>resistor value. I have googled the problem and come up short. Anybody got
>>a pointer to a URL that goes through the math of how this configuration
>>works? And what I'm doing to my phase margin?
>>
>>Jim
>
> (Refer to Jim Thompson's schematic in alt.binaries.schematics.electronic)
>
> In what follows, I'll assume the capacitor is replaced with a short and only first order
>effects will be considered.
>
> An easy way to see what is happening in the "T" (multi path) network feedback
>arrangement is to transform the "T" to a "Pi" network.
>http://encyclopedia.thefreedictionary.com/Y-delta+transform
>
> First, consider what happens in an inverting amplifier of the sort in Jim's schematic
>when you connect a resistor from the "-" input of the amp to ground. The "signal" gain
>remains unchanged, but the "noise gain" of the circuit changes.
>
> Now apply the well known "Y-Delta" transformation. Jim shows a "T" with (left to right;
>input to output side) a series arm of 9000 ohms, a shunt arm of 1000 ohms, and a series
>arm of 99200 ohms. This is equivalent to a pi network (left to right) with a shunt arm of
>10090.7 ohms, a series arm of 1.001 megohms, and a shunt arm of 111222 ohms. The 10090.7
>ohm shunt arm on the left goes from the "-" input to ground and so has no effect on the
>signal gain. The 1.001 megohm series arm provides essentially the same feedback as the
>1.000 megohm resistor in the two resistor amplifier circuit. The 111222 ohm shunt arm
>from output to ground is simply an additional (small) load on the amplifier and has no
>effect on circuit performance.

Why hasn't someone pointed out that by exchanging the left and right sides of the T
network, we could have a shunt arm on the minus input of 111222 ohms, instead of 10090.7
ohms, a factor of 10 better. The 10090.7 ohm arm will then be a load on the amplifier
output, still without effect. The series arm is unchanged of course, so the net effect
may be an improvement. A possible undesirable consequence might be that the input
capacitance of the op amp might be greater than the stray capacitance at the junction of
the 3 T-network resistors, so with the reversed arrangement the op amp capacitance is fed
with 99200 ohms, rather than 9000 ohms. One would have to check this possibility.

Why didn't *I* point this out before now?

>
> The noise gain in the two-resistor circuit is 1 + R3/R1; in the multi path feedback
>circuit, the noise gain is 1 + R3/R1 + R4'/R1, where R4' is the 10090.7 shunt arm in the
>equivalent delta network. Since R4' is about .01 times the value of R3, we should expect
>no easily observable low-frequency difference in the performance of the two circuits, as
>Jim's simulation shows. The noise gain is also the gain applied to the op-amp offset
>voltage, and the capacitor in the multi path circuit prevents the increase in this gain at
>DC. But in this particular multi path circuit, the increase in noise gain is only around
>1%, so we could omit the capacitor.
>
> This method (using the Y-Delta transformation) of analyzing the circuit indicates to me
>that Jim's assertion:
>
>"...the 1:1 case has no useful value toward improving (lowering)
>the total impedance in the feedback loop."
>
> is incorrect. Consider the following:
>
> In Jim's multi path circuit, make R4 9803.92 ohms, make R6 98.0392 ohms, and make R5
>9803.92 ohms. Now the Pi equivalent network has the two shunt resistors equal to 10000
>ohms, and a series arm equal to 1.000 megohms, which gives a (low-frequency) circuit
>performance identical to the two-resistor circuit, except for about 1% higher noise gain.
>The total impedance in the feedback loop is much lower than the two-resistor circuit.

From: Fred Bloggs on


Terry Given wrote:
> Fred Bloggs wrote:
>
>>
>>
>> Terry Given wrote:
>>
>>>
>>> I recently built about 50,000 of this circuit, with a feedback cap
>>> too (mathcad rather than mathematica, and a pencil to start with for
>>> the analysis), and 15 inputs thru 100k resistors. the effect of the
>>> 14 "grounded" resistors shifted the center frequency by about 10% -
>>> Aol was about 50. power consumption (and cost) constraints meant I
>>> couldnt use a faster opamp, so instead I stopped assuming and started
>>> calculating :)
>>>
>>
>> What was the transfer function you were shooting for, and which amp?
>>
>
> a summing band-pass (ish) filter. 40 x TLV274.
>
> I didnt want to AC-couple the inputs (that would have cost me 240
> capacitors) so I used the bridged-T feedback network with an RC shunt to
> give a DC gain of about 1/16 - any DC is basically common-mode, and the
> next stage was AC coupled. 3 Rs and 2 Cs was a whole lot cheaper than an
> RLC. But 100k/14 = 7k in parallel with the -ve shunt arm, enough to move
> Fc 10% or so.
>
> SPICE clearly showed it, so I went back and re-did my opamp analysis
> using Dostals approach (originally I did it using the Woodgate
> approximation), and voila - out popped the same answer. Mr HP3577 also
> agreed with spice and mathcad. Dostals method also allowed me to
> directly calculate the phase margin. Since then, I have analysed all
> opamp circuits thusly - but I use the Woodgate approach with pencil &
> paper as a bullshit detector :)
>
> Cheers
> Terry

You can achieve a wild increase in effective GBW by going to current
mode feedback. The peaking and rapid rolloff due to that low impedance
-ve shunt is eliminated from any frequency bands usable with the voltage
feedback circuit. The output gain of 2x deals with the CMR input range
of the TLV274- requires about a volt of headroom to V+ - facilitation
odds and ends not shown...
View in a fixed-width font such as Courier.

..
..
.. >--[Ri]-+-------+------[R1]--+--[R2]------------+
.. | | | |
.. o | | [R3] +--[R]---+-->Vout
.. | | +5V | | |
.. | | | C | | |\ |
.. o | | +-----||-+ +--|-\ |
.. | | | | | >--+
.. | | +-----[Rc]-----+------|+/
.. o | | | | | |/
.. | | | | |
.. >--[Ri]-+ | +-----------+ | [R]
.. | | | | | |
.. >--[Ri]-+ +------|>|---+ | | |
.. | | | | | | |
.. >--[Ri]-+ +--------------------------+
.. | | | | | |
.. | | | | | |
.. | | |\| c | |
.. | +-|+\ |/ | | OA TLV274
.. | | >-+---| c |
.. 2.5V>-+-----|-/ | |\ |/ |
.. |/| | e+ c
.. | | |\ |/
.. | [Rb] e+
.. | | |\
.. | | e
.. | | |
.. GND---------+--+-----------+------------
..
..
..
..
..
..
..
..
..
..
..
..

From: Terry Given on
Fred Bloggs wrote:
>
>
> Terry Given wrote:
>
>> Fred Bloggs wrote:
>>
>>>
>>>
>>> Terry Given wrote:
>>>
>>>>
>>>> I recently built about 50,000 of this circuit, with a feedback cap
>>>> too (mathcad rather than mathematica, and a pencil to start with for
>>>> the analysis), and 15 inputs thru 100k resistors. the effect of the
>>>> 14 "grounded" resistors shifted the center frequency by about 10% -
>>>> Aol was about 50. power consumption (and cost) constraints meant I
>>>> couldnt use a faster opamp, so instead I stopped assuming and
>>>> started calculating :)
>>>>
>>>
>>> What was the transfer function you were shooting for, and which amp?
>>>
>>
>> a summing band-pass (ish) filter. 40 x TLV274.
>>
>> I didnt want to AC-couple the inputs (that would have cost me 240
>> capacitors) so I used the bridged-T feedback network with an RC shunt
>> to give a DC gain of about 1/16 - any DC is basically common-mode, and
>> the next stage was AC coupled. 3 Rs and 2 Cs was a whole lot cheaper
>> than an RLC. But 100k/14 = 7k in parallel with the -ve shunt arm,
>> enough to move Fc 10% or so.
>>
>> SPICE clearly showed it, so I went back and re-did my opamp analysis
>> using Dostals approach (originally I did it using the Woodgate
>> approximation), and voila - out popped the same answer. Mr HP3577 also
>> agreed with spice and mathcad. Dostals method also allowed me to
>> directly calculate the phase margin. Since then, I have analysed all
>> opamp circuits thusly - but I use the Woodgate approach with pencil &
>> paper as a bullshit detector :)
>>
>> Cheers
>> Terry
>
>
> You can achieve a wild increase in effective GBW by going to current
> mode feedback. The peaking and rapid rolloff due to that low impedance
> -ve shunt is eliminated from any frequency bands usable with the voltage
> feedback circuit. The output gain of 2x deals with the CMR input range
> of the TLV274- requires about a volt of headroom to V+ - facilitation
> odds and ends not shown...
> View in a fixed-width font such as Courier.
>
> .
> .
> . >--[Ri]-+-------+------[R1]--+--[R2]------------+
> . | | | |
> . o | | [R3] +--[R]---+-->Vout
> . | | +5V | | |
> . | | | C | | |\ |
> . o | | +-----||-+ +--|-\ |
> . | | | | | >--+
> . | | +-----[Rc]-----+------|+/
> . o | | | | | |/
> . | | | | |
> . >--[Ri]-+ | +-----------+ | [R]
> . | | | | | |
> . >--[Ri]-+ +------|>|---+ | | |
> . | | | | | | |
> . >--[Ri]-+ +--------------------------+
> . | | | | | |
> . | | | | | |
> . | | |\| c | |
> . | +-|+\ |/ | | OA TLV274
> . | | >-+---| c |
> . 2.5V>-+-----|-/ | |\ |/ |
> . |/| | e+ c
> . | | |\ |/
> . | [Rb] e+
> . | | |\
> . | | e
> . | | |
> . GND---------+--+-----------+------------

I'll study that a bit later. unfortunately it also achieves a wild
increase in parts count and cost - this circuit is replicated many, many
times :)

Plus of course the original works just fine, and is in production.

Cheers
Terry