[R-390] Comments on ER Issue 208 AGC modification

[Charles Steinmetz]

Many, many moons ago I said I would post an analysis of the ER AGC mod 
(Electric Radio, Issue 208) if I saw the article.  A helpful list member 
sent me the article some time ago, so here goes.  It's long, and pretty 
technical, so grab a cuppa joe and a pile of schematics before trying to 
digest it.

First, I'll go through the stock AGC circuit (including brief 
descriptions of the mods I do when I do simple AGC modifications), then 
proceed to the ER-208 mod.  Readers will need to have a schematic of the 
R390A at hand to follow the first part, plus the mod article from 
Electric Radio, Issue 208 to follow the second part.

AGC OVERVIEW:
The purpose of AGC in the 390A is to develop a negative voltage that 
varies with the IF signal level, which is used to bias the RF stage, the 
mixers, and the IF stages.  By varying the bias of these stages, the RF, 
mixer, and IF gains are changed (they all have less gain when the AGC 
voltage is more negative).  The AGC voltage is derived by rectifying and 
low-pass filtering (quasi-integrating) the IF signal, much like AC from 
a power transformer is rectified and filtered in an AC-DC power supply.

In the stock circuit, an IF signal is taken from the IF cathode 
follower, V509B, and amplified by the AGC IF amp, V508.  The amplified 
IF signal is fed through a 220 pF capacitor, C546, to diode clamp V509A 
(called the "AGC rectifier").  V509A clamps the V509A end of C546 to a 
potential near ground when the plate of V508 (and its low-Q tuned load, 
Z503) swings positive, and C546 charges up as its V508 end goes positive 
until the plate reaches its peak positive voltage (about 200v).  After 
the IF signal on the plate of V508 crests and starts back down, the 
V509A end of C546 is no longer diode-clamped to ground by V509A, so the 
charge in C546 drives the 100k resistor R545 negative, following the 
voltage on the plate of V508 but some 200-odd volts less positive.

This happens on every cycle of the 455 kHz IF signal, and the end result 
is that a half-wave rectified version of the 455 kHz IF signal appears 
across R545, with the positive excursion clamped near ground by V509A 
and the half-wave pulses extending negative from ground.  Using C546 
this way is a form of "charge pumping" -- C546 is charged on each 
positive half-cycle of the 455 kHz IF signal, and some of that charge is 
transferred or "pumped" (with integration due to R546) to C547 on each 
negative half-cycle.  The clamped, half-wave rectified IF signal on R545 
is fed through 180k resistor R546 to 0.1uF filter capacitor C547, to 
produce a DC voltage that varies with the received signal strength 
(around +12v with no signal input [see below re: R544], and as much as 
30v negative with strong signals).

IMPORTANT NOTE:  The analysis below assumes that V508 and its plate load 
(Z503) are able to source and sink more current than necessary to charge 
the AGC capacitors at the time constants listed.  This is NOT always 
true in practice, so the actual attack times observed can be 
SIGNIFICANTLY LONGER than the calculations indicate.  This is discussed 
in more detail below.

If there were no IF signal, the non-grounded end of C547 would sit at 
around +12 VDC due to the voltage divider from B+ formed by 2.7M 
resistor R544 and 180k resistor R546 to the anode of the diode clamp, 
V509A.  This positive bias is countered (and overcome, with any 
appreciable IF signal) by the negative-going, half-wave rectified IF 
signal flowing through R546 into C547.  The time constant to charge C547 
through R546 is around 18 mS (T=RC; 180k x 0.1uF = 18 mS).

SIDE NOTE: the "time constant" ("T" = RC) is NOT the time it takes to 
fully charge or discharge the capacitor to the new voltage -- it is a 
mathematical construct used for analysis and is related to frequency as 
expressed in radians.  A capacitor fed by a resistor charges to 63% of 
its new value in 1 x T (= 1 x RC).  After 2 x T (= 2 x RC), the 
capacitor voltage reaches 87% of its new value, and after 5 x T (= 5 x 
RC) it reaches 99% of its new value.

The time constant to DISCHARGE C547 is more challenging to calculate; 
the discharging resistance is 280k (R546 + R545), which would give T = 
RC = 28mS; BUT the discharge is assisted by the bias current through 
R544 (which is close enough to a constant current of 70uA to be treated 
as such).  A constant current charges (or discharges) a capacitor at a 
constant linear rate R = dV/dT = I/C.  For I = 70uA and C = 0.1uF, R = 
700 volts per second (= 0.7 v/mS).  So, the effective discharge rate is 
significantly faster than T = 28mS, and depends on how far (how many 
volts) the AGC needs to go from its old value to its new value.

The AGC filter is actually two sections -- one comprising R544, R545, 
and R546 along with C547, discussed above, followed by another section 
comprising 220k resistor R547 and some combination of capacitors C548 
(0.1uF) and C551 (2uF).  The AGC Line is fed from the non-grounded end 
of C548.

For Fast AGC, C551 is out of the circuit and the second section 
comprises only R547 and C548.  The time constant of this combination by 
itself is 22ms, although that is only an approximation because the end 
of R547 toward C547 is not grounded for the low frequency AGC signal. 
Together, the two sections give a two-pole response with an effective 
time constant ("T" = RC) of around 35mS for both charge and discharge 
when Fast AGC is selected.

When Medium AGC is selected, C548 is paralleled by 2uF C551.  In this 
case, C547 does next to nothing and the time constant ("T" = RC) is 
approximately symmetrical (charge = discharge) with a single-pole value 
of something over 400 mS.

Finally, when Slow AGC is selected, C548 is again paralleled by C551, 
but this time the "far end" of C551 (the end away from C548) is not 
attached to ground, but rather to the plate of AGC time constant tube 
V506A.  Because C551 is connected from the plate of V506A to its grid, 
its value is increased by the Miller effect.  In this case, the 
effective value of C551 is around 20uF, giving a single-pole time 
constant ("T" = RC) of around 4 seconds.

BONUS!!   EXPLANATION OF THE DREADED "MOMENT OF SILENCE":
Note that the end of C551 away from C548 gets switched to ground when 
Medium AGC is selected, and to the plate of V506A when Slow AGC is 
selected.  Because the plate of V506A is not at ground potential (it 
sits at around +30 VDC when there is little signal, and can reach ~ 
+200VDC under strong-signal conditions), the very large C551 must charge 
and discharge into the AGC network every time you switch between Slow 
and Medium AGC.  This is why the radio goes silent for a while when you 
switch from Slow AGC to Medium AGC!  When you throw the switch, the AGC 
line goes immediately to about 30 volts more negative than it was, and 
you have to wait while it discharges through R547, R546, R545, and R544. 
  Because this series string has high resistance, the discharge (AGC 
recovery) is slow (several seconds).  [The reverse happens when you 
switch from Medium to Slow AGC -- there is a period of distorted audio 
from the overloaded IF sections.  This period is very brief, however, 
because the capacitor discharges through the following grid-cathode 
diode, which has a much, much smaller resistance than the series string 
of R547, R546, R545, and R544.  It generally passes unnoticed.]

To cure this -- and only this -- you can very easily modify the AGC 
circuit by adding a 20uF film capacitor to ground and permanently 
grounding the end of C551 that is away from C548, so that the AGC switch 
adds C551 in parallel with C548 for Medium AGC and the new 20uF film 
capacitor in parallel with C548 for Slow AGC.

Now, apart from the "muting after switching" or "Moment of Silence" 
problem, why would one want to modify the R390A AGC system?  Primarily 
because symmetrical attack and release time constants (equal time to 
charge and discharge the AGC capacitance) are not the hot setup.  We 
generally want our AGC to attack substantially faster than it releases. 
  To do this, one needs to charge the AGC filter capacitor(s) from a 
lower resistance than the discharge resistance.

Another reason one might want to modify the R390A AGC system is to apply 
AGC differently to the various gain-controlled stages to maximize 
headroom and minimize noise.  As a general matter, this is known as 
"staged AGC."  In the particular case often encountered in radio AGC 
circuits, it is commonly known as "delayed AGC."  NOTE that in this 
usage, "delayed" does not (except possibly incidentally) mean "delayed 
in time" -- it means "delayed with respect to the IF signal level."  In 
other words, AGC in NON-delayed stages is directly proportional to the 
IF signal level, while AGC in DELAYED stages is proportional to the IF 
signal level ONLY ABOVE A THRESHOLD (the "delay threshold").

THE "TWO-DIODE" AGC MODIFICATION:  [The "two-diode" AGC mod is often 
(erroneously, in my view) referred to by the name of one person who 
wrote about it, but it had been widely used for decades before that.] 
One very easy way to produce asymmetrical charge/discharge time 
constants in an R390A is simply to parallel R546 and R547 with 
solid-state diodes (1N914/1N4148 or equivalent), with both cathodes 
toward V509A.  This arrangement turns C546 into a classic charge pump 
(with no integration due to R546 and R547).  I have installed this 
modification into scores of R390As since the early 1960s (along with 
adding the 20uF capacitor and switching changes as discussed above to 
eliminate the "muting after switching" problem), with excellent results. 
  (I sometimes change the values of C548, C551, and the new capacitor, 
and/or the value of R547, to tailor the AGC action for particular uses. 
  I also sometimes put some resistance in series with the diodes to slow 
down the attack time.)

NOTE, HOWEVER, that this method cannot produce extremely fast attack 
times in any event, because Z503, V508, and the C546 charge pump can 
only supply limited current to charge the AGC capacitor(s).  To achieve 
really fast attack times, particularly with Medium and Slow AGC, you 
need to add a current amplifier between the plate of V508 and C546.  I 
have used solid state emitter (BJT) or source (FET) followers for this, 
for a much improved AGC attack response.

To achieve a 1mS attack time with Slow AGC, the AGC dectector needs to 
supply 400 mA to charge C551 (!!).  The current to charge the AGC filter 
capacitors ultimately comes from the AGC IF amplifier V508, a 5749, 
which can deliver only 12-15 mA on a good day.  So the best you can do 
without adding a current amplifier is a 40 or 50 mS attack time in Slow 
AGC, and 4 or 5 mS in Medium AGC.  [NOTE: These figures are NOT the time 
constant "T" (= RC).  They are the full-scale attack times, more like 3 
x T.  See the "SIDE NOTE:" above.]

THE ER 208 MODIFICATION:
With that background, let's turn to the ER mod (have the ER article 
handy as you read the following):

The author of the ER article (whom I do not know) used a solid-state 
diode to rectify/clamp the 455kHz IF signal, and re-purposed V509A as an 
additional, common-cathode AGC IF amplifier.  The plate of V509A feeds a 
510pF charge-pumping capacitor, and solid-state diode D1 performs the 
clamping function originally performed by V509A.  In this case, the 
half-wave rectified IF signal is clamped not to ground, but to the 
cathode of V509A at a stated +5.2 VDC.  Presumably, this is to give the 
raw rectified IF signal a positive bias, as was accomplished in the 
original circuit by the 70uA current through R544 (see above).

The current that charges the 510pF pumping capacitor on the 
positive-going swing of the V509A plate flows through D1 into the 1k 
cathode resistor and its 0.01uF bypass capacitor, and will therefore 
cause the bias on V509A to wander with the IF signal level.  The 0.01 uF 
bypass capacitor begins to effectively bypass the V509A cathode at 
around 15 kHz.  In other words, audio frequencies are NOT effectively 
bypassed, while the 455 kHz IF frequency IS fairly effectively bypassed. 
  Therefore, if the IF level varies significantly at audio or sub-audio 
frequencies -- which it will not do very much with an AM signal, but 
most certainly will with an SSB or CW signal -- there will be positive 
cathode feedback in the V509A AGC IF amplifier at audio frequencies. 
This may explain why the author found it necessary to attenuate and 
further integrate the IF voltage to the RF amplifier for undistorted CW 
reception (with new R1, R2, and C3, T = RC = ~100mS).  It would be much 
better design practice to obtain the positive AGC bias some other way.

V509A and its 5.6k plate resistor can supply about double the current to 
the charge pump than V508 and Z503 can in the stock circuit, so we 
expect that somewhat faster attack times are possible with the modified 
circuit.  During the negative-going portion of each cycle of the V509A 
plate signal, the charge put into the 510pF capacitor during the 
positive-going portion of the cycle is pumped through D2 to charge the 
various AGC filter capacitors.  Because one of the filter capacitors 
(the 0.005uF capacitor) is charged directly by the 510pF pump capacitor 
through D2, the AGC loop can have a very fast attack time.  However, 
even in Fast AGC mode, most of the filter capacitance -- 0.1uF -- is fed 
through a 47 k resistor, so the bulk of the AGC is integrated with a 
time constant (T = RC) of ~4.7mS.  This means that for transient events 
(faster than a few mS), the AGC can attack very quickly but will also 
release quite quickly (T = RC = ~235uS) as the charge from the 0.005 uF 
filter capacitor flows through the 47k resistor into the 0.1uF filter 
capacitor.

This characteristic may be useful for reception during lightning and 
other transient disturbances, although for this purpose I would be 
inclined to increase the value of the 0.005 uF capacitor a bit and 
consider reducing the value of the 0.1uF capacitor similarly.  The 
Medium and Slow AGC capacitors each have 100k resistors in series, 
limiting the attack time constants ("T") to 50mS and 100mS, 
respectively.  I would be inclined to reduce those resistors to 39k 
(Medium) and 20k (Slow) to give T = 20mS for both.  In this case, the 
full scale attack time is about 3 x T, or 500 mS and 1 S for the 
author's values and 200 mS for my suggested values.  [NOTE that even 
these values are MUCH too slow, IMO.]

Unlike the stock circuit, which discharges the AGC filter capacitor(s) 
through R547, R544, R546, and R545, the modified circuit discharges the 
AGC filter capacitor(s) through new R1 and R2.  With the values given in 
the schematic, the discharge (AGC decay) time constant ("T") in Fast AGC 
mode is about 150mS.  In the text, the editor states that he used 680k 
for R2 and sometimes used 270k for R1.  With those values, the Fast AGC 
decay time constant ("T") is around 100 mS.

For Medium AGC, the attack time constant (T = RC) is about 47 mS and the 
release time constant is about 700 mS (schematic values) or 500 mS (with 
the ER editor's values).  For Slow AGC, the attack time constant (T = 
RC) is about 100 mS and the release time constant is about 1.5 seconds 
(schematic values) or 1 second (editor's values).  In both cases, the 
Fast AGC components provide additional fast attack with fast release on 
transients under ten milliseconds, as discussed above.  In my view, a 
release time constant of 1.5 seconds is much too fast for Slow AGC, so I 
would be inclined to increase C1 to at least 2 uF, and, better, to 4.7 
uF or possibly even 10 uF, depending on one's choice of R1 and R2.  The 
resistor in series with C1 could be reduced to 10k or even lower to 
maintain the attack time closer to 20 mS, but only so much can be done 
because we run into the finite ability of V509A and its 5.6k plate 
resistor to supply current to the charge pump.

I noted above that a second reason to modify AGC systems could be to 
apply AGC differently to different stages in order to optimize headroom 
and noise.  As noted, this often takes the form of "delayed AGC." 
(Again, "delayed" here does not mean delayed in time; rather, it means 
delayed with respect to the RF input level.)  As the input signal level 
is gradually increased from 0, the gain of some stages is reduced 
significantly before the gain of other stages is affected at all.  By 
attenuating the AGC feed to the RF section, the ER modification provides 
quasi-delayed AGC to the front end.  By "quasi-delayed," I mean that it 
isn't "delayed" in the usual sense, where the stages that have delayed 
AGC run at full gain until a certain input level (the "delay 
threshold"), then start reducing gain as the input further increases.

Rather, this "quasi-delayed" AGC starts immediately (with increased IF 
level), but because it is attenuated, the gain of the "quasi-delayed" 
stages doesn't go down AS FAST with increased RF input as it did 
originally, so the RF stage never does get fully gain-reduced (or at 
least not until the IFs are fully cut off).  I see no reason to believe 
that this arrangement optimizes either headroom or noise.  In fact, 
there is good reason to believe it is the OPPOSITE of what one would 
want.  So, my intuition is that this is far from an optimal solution, 
given the known overload problems in the R390A front end.  But one would 
  need to do gain, noise, and headroom analyses of every stage from the 
antenna input to the detector to know for certain.

Best regards,

Charles