Using feedback in AA5 tube radio audio amps for better fidelity
Fig 1: 
Fig 3:
Fig 1:
Fig 3:
If you have a particularly good sounding AA5 radio with a nice speaker and cabinet (or you could replace the speaker with one intended for car audio), it is possible to add some negative feedback for better fidelity. Some reduction in hum can also be had with negative feedback. There's usually plenty of audio gain avaliable in an AA5 tube radio to allow some use of feedback, as feedback will reduce gain. At first I tried some feedback into the cathode of the 12AV6, but this cathode also serves the diode detector circuit. Oh, it worked, but the feedback ends up being applied to two paths, which, depending on the volume control setting, would tend to cancel out. One solution to this problem is to use a twin triode tube like a 12SL7 or 12AX7, using one section as the triode audio stage with the feedback going into the cathode. Place an additional 56K resistor from the plate to B+ to make the plate load 47K. And use a 2K resistor on the cathode, this cathode resistor bypassed to the voice coil output transformer secondary. Doing this gives us about 2 dB of additional gain, as the plate impedance is lowered. Thus the output tube grid resistor won't load it as much. I used a surface mount resistor on the bottom of the AA5's circuit board. The other triode section is wired as a detector diode, grid to the IF transformer, plate and cathode to ground. See Fig 2 below.
Another method without this drawback is possible. It involves adding a third winding to the audio output transformer. About ten turns around one of the outer legs of the iron core laminations will do. See Fig 1. But as this winding will be carrying the audio input signal, it will need to be shielded from the other windings of the transformer. Use some copper foil tape between this new winding and the other windings of the transformer. And also use some foil tape to shield the windings from the outside world. Ground these shields and the transformer laminations. Be sure not to create a shorted turn around the leg of the iron core. And shield the two leads of this new winding from the transformer back to the 12AV6 circuit. Or if there's enough room in the transformer, you could use very thin coax cable for the new windings. Ground only one end of the shield of the coax. In Fig 3, I didn't need the foil tape. I grounded the transformer laminations and also grounded the voice coil. The voice coil winding (being the outer winding over the primary winding) being grounded acts to shield the new feedback winding from the primary winding. This feedback winding subtracts a little bit of the output signal from the input signal without affecting the detector circuit. That then creates the negative feedback loop. If phased properly, you should be able to short the feedback winding and hear the audio level *increase* a couple of dB, and have less fidelity.
The capacitor usually connected to the plate of the output tube can be reduced to a few hundred pF's, or removed completely. The feedback loop will now keep the radio from sounding shrill.
Which way is the audio transformer phased? We need a negative feedback loop. Phased the wrong way we will get an oscillator instead of an amplifier! If that happens, reverse the leads on the audio output transformer's new winding at the 12AV6 (Fig 1), or the speaker driver winding if you used feedback into the first audio triode's cathode (Fig 2).
For more bass, install a 0.1uF cap from the plate of the triode driver stage to the output tube's control grid. But don't install a bigger cap on the grid of the triode driver to the volume control, as it could back bias the diode detector more, see below.
Fig 2: 
Modulation acceptance is the amplitude modulation percentage (aka modulation index) a receiver's detector can handle without distortion. An ordinary diode detector can handle upward modulation well but detector circuits with capacitive loads have limited ability to faithfully reproduce downward modulation. At some point they clip the audio waveform before getting to zero. Typically an AA5 will start to clip modulation that drops below about 20%. FCC rules say the minimum is about 5%. This level is when the music is at the loudest. The RF carrier varies between 5% and 195% of the "dead air" RF level when the music is loud. This is known as "95% modulation". AA5s tend to clip on modulation above the high 80's.
An easy way to improve this situation is to make use of the second diode in
the 12AV6 or 12SQ7 tube. Disconnect the AVC resistor from the volume control
or the IF transformer and then connect it to the anode of the extra diode. Add
a new 470K resistor to that diode's anode and the other end to the 12AV6 cathode
(usually ground). And finally connect a 30pF cap between the plate of the IF amp
tube and the newly used 12AV6 diode anode. See diagram below. This gives us
AVC action without messing up the detected audio, as they are now separate
circuits. This requires two extra parts in an AA5, something manufacturers
would avoid in a mass market consumer product. This mod may not make much
difference in a small table set, but it should greatly improve a large
console set or the AM section of a stereo system.
The audio coupling cap from the volume control wiper to the grid of the
12AV6 can also add to the problem, but the voltage divider effect between the
wiper resistance to ground and the 10 meg resistor pushes it down to a per-cent
or so. And this cap, being smaller, would discharge faster.
A cathode follower could be used as a buffer to avoid even this, but
the cost would preclude its use except in a very high end set.
Reduced heater voltage on detector diodes like the 6H6, 6AL5 or 5896
can improve detector performance on weak signals. Less "contact
potential" for the signal to overcome. This should increase the
fidelity of AM detection. The 5896 below is a dual diode version.
This distortion improvement is due to a better AC/DC load ratio
on the detector. The DC load is the resistance directly connected to
the detector. The AC load is the DC load with the addition of the
resistance on the other side of the coupling cap feeding the grid
of the triode.
To avoid "negative peak clipping" you want the
AC/DC load ratio to be as close to 1.0 as possible. The further the ratio
is from 1.0, the lower the modulation level will be where "negative peak
clipping" sets in. The actual negative modulation percentage where
negative peak clipping starts is determined by the source impedance of
the IF stage driving the detector, and the diode characteristics, as well
as the AC/DC load ratio.
For comparison a typical "AA5" radio has a DC detector load of 547K
and a worst case AC detector load of 433398 Ohms with the Volume
control at maximum, for a AC/DC load ratio of 0.792.
With the change of the detector load from 500K to 83K (100K in parallel
with the 500K pot),
the DC load becomes 83K, and the AC load becomes 81K.
The resulting AC/DC load ratio of 0.975 is a considerable improvement over
the AC/DC load ratio of 0.792 as originally designed. (John Byrns, with edits)
The
signal level of each circuit will drop about 6 dB however.
You might
be able to get some gain back by connecting a small 15pf or
so cap from the hot side of the primary to the hot side of
the secondary of the IF transformers. This depends on the
phasing of the magnetic coupling inside the transformer, however.
The antenna circuit's "Q" can also be lowered. In this
circuit, one would like an approximately constant bandwidth
over the range of the AM MW band. A small resistance in
series with the antenna coil before it connects to the tuning
capacitor and converter tube will do this. Install a 27 ohm
resistor here. After all this, you'll find the radio will
only hear the local 50 thousand watt flamethrowers in town.
But with better audio response. You can try to boost the gain
in the IF stage by bypassing the cathode resistor with a 0.1uF
cap to ground. You can use a low voltage cap here, as there
will only be a volt or two across it here.
Something to watch out for is 10KHz whistles caused by out of
town station carriers if the bandwidth gets too wide.
If you have the typical modern digitally tuned AM/FM stereo receiver
for your home audio system, you probably noticed the poor quality
of the audio from the AM section of the tuner.
No audio high frequencies at all (above about 4KHz). As stated above,
AM stations broadcast audio up to 10KHz. Which makes their AM
modulated signal have 20KHz bandwidth. The FCC assigns carrier
frequencies further apart than this in your particular town. Out of
town signals on adjacent channels are usually too weak to be heard
on your local station.
Most modern receivers use a ceramic filter of about 10KHz at most,
yielding audio that tops out at 5KHz. What I did to a set that
uses a Sanyo LA1851N AM/FM stereo chip is remove the AM
ceramic filter and its IF and replace it with a set of 3
IF interstage transformers taken
from
an old narrow band FM pager transistor radio (GE model 4er35a12 if you happen
to have such laying around). See diagram below.
All you need are 3 interstage IF transformers and a pair of 4pF caps.
Here I cut out the section of the old pager circuit board that has these
IF transformers and caps already wired and wired it to the stereo
receiver board where the old ceramic filter and its IF was. You
could do a neater installation than this, but keep the signal wires
as short as practical.
Stations sound much
better now, I get most everything they transmit. This is a
spectral plot of the detected audio from a local AM station. This was
taken from an "S" sound from voice audio. Reasonably flat to
10KHz, the station's transmitter NRSC brickwall low pass filters
to
A simulation of the 3 IFs:
And the group delay, which is symmetrical. 20usec differential group
delay has no real impact on the demodulated audio anyway.
If the set
is digitally tuned, you must select a new filter to be the same
IF frequency as the old one. Or else the set won't tune on channel
correctly. They come in 450, 455 and 460Khz.
If the new filter is off by 10KHz then the set will
tune stations one 10KHz channel off, but otherwise sound fine.
The tracking across the band will be slightly off.
Worse yet is a filter 5Khz off. Then you can never tune the
station right.
If the set is analog tuned (ie, slide rule or round dial with
twist knob, with an old fashioned tuning cap) the worst that happens
is that the dial calibration will be off slightly and tracking off slightly.
Most dials
aren't this accurate anyway.
I used a pair of 5906 subminiature sharp cutoff pentodes. Triode
strapped, but is that significant in cathode follower service?
Well, I had lots of 'em, also the 26. 5V heaters makes it easier
to run off the SS power amp power supply (+65V and -65V). Also submini
tubes are "cute", and small enough to shoehorn inside this receiver.
Used the +65V supply with series resistor and with the two tube
heaters in series. Okay, but how about B+? Well, I built a kind of
voltage trippler circuit with a second bridge rectifier and some "AC
coupling" caps. See circuit. I built this on a salvaged small
switching power supply board, using its old AC line bridge rectifier
and filter cap. This board made a handy way to mount this circuit.
The coupling caps went where an inductor filter used to live.
The diagram below shows a simulated amount of power supply ripple
(which is the amount I saw with a DVM via a cap after I built the
power supply with a dummy load, but before I built the cathode
follower circuit), and the tube's power supply rejection on the
output. Which is pretty good.
The B+ is just directly rectified off the powerline, making this a hot chassis
radio. But this radio chassis is easily isolated for safety.
The curves below show the 5639 in pentode mode, standard triode connection mode, and
the modified triode connection, with the 5K resistor between the plate and screen,
and the load taken off the plate and the 5K resistor node. Any load line centered
at 150V and 30ma drawn on the pentode mode curves or the standard triode connection curves
will have rather severe non-linearities. The modified triode mode has
significantly more linearity here.
Assuming that the signal level at the detector is high enough
to avoid the "knee" (non-linear region just above zero volts, which if you are
using a 6AL5 can be reduced by running its heater at 4V) of the detector diode
and only the use of RC filter caps to remove the IF frequency from the detected
audio, you can get around -50dB distortion products in the audio out. This RC
filter is the 100pF caps and the 47K resistor. If you then connect an AVC
filter network to the audio output, the 0.05uF cap will take longer to discharge
than the period of time of the typical audio waveform. Even though the AVC resistor
(usually 2.2 or 3.3 megs) and the volume control (usually 500K) form a voltage divider
for the voltage stored in the AVC cap, about 20% of the AVC voltage shows up at
the detector diode. This causes the diode to be back biased, requiring extra
voltage from the AM signal to conduct. This has the effect of clipping off
the bottom excursion of the AM signal's carrier's modulation.
Another thing you should address is the small RF filter caps on the audio
output of the detector. Have them too big and you can get tangent distortion
because they take too long to discharge to allow a full amplitude 5KHz sine
wave thru without getting tangent clip. Change the (usually) 100pf
caps to around 33 to 47pF.
This reduced heater trick seems to also work well on 12SQ7's and 12AV6's.
Only thing is to keep in mind that if you reduce the contact potential
for the diodes, it also drops for the triode. But it seems that running
the heater at 10V vs 12.6 seems to be a sweet spot. Parallel a 330 ohm
half watt resistor with the heater in a series string set. If it's a
6SQ7 in a 300ma heater string, use 160 ohms if you have it, or use
150 ohms. It shouldn't hurt the tube, as the current demand on the
triode in an AA5 is quite low. The below graph (original source before edits:
http://www2.famille.ne.jp/~teddy/datalib/heater.htm) shows the impact of
reduced heater voltage on the 12AV6 triode. At 10V heater voltage the
plate current curves are slightly shifted to lower plate currents
vs plate voltage, but not adversely so. Around 9V things start to
fall apart.
Even better AM detector using cathode followers and vacuum tube diodes
For a "hi-fi" AM detector that can handle a wide range of signal
levels, and where you are willing to use extra tubes, a detector that is
double buffered by cathode followers works well. Distortion down about 50dB
and about 40dB for really weak signals (other detectors destroy weak signals).
The first cathode follower buffers the radio station carrier from the last IF transformer.
This allows heavy loading by the detector. Then there is a pair of detector vacuum
tube diodes, one for the AVC, and the other for audio extraction. Then the recovered
audio is directly coupled to a second cathode follower. This keeps to an absolute
minimum the capacative loading of the detector circuit. Especially no coupling capacitor
to back bias the detector diode. With C3, R4, C2 and R5 carefully selected for
a good AC/DC ratio, this circuit can handle input signal levels from 20Vp-p to as
low as 1mVp-p, with distortion levels of -50dB to -40dB respectively. The second cathode follower
needs a negative supply to keep it linear on larger audio signals. I obtained -8V
by rectifying the 6.3VAC heater supply in the Heathkit tuner I modified with this
circuit. Another way is to bias the audio detector circuit's ground up to around +25V.
This gives plenty of headroom for the 2nd cathode follower to handle large signals.
Once the simulations looked good I built hardware, and it sounds very good.
The IF transformer secondary feeding the first cathode follower has a 100K resistor
added across it to widen its passband now that the detector load isn't on it
(not shown in the below diagram).
At the expense of sensitivity, the audio bandwidth response
of an AA5's RF and IF can be improved. This involves lowering
the "Q" of the tuned circuits in the antenna and IF strip.
Install a 100K resistor across the secondary of the first
IF transformer.
This should lower the "Q" to
a value that will yield about 15 to 20KHz bandwidth.
The secondary of the second IF transformer will be loaded
with an extra 100K load on the detector output. This should
improve distortion as well as lower the transformer's "Q".
The AM NRSC-1 standard is:
Contrary to popular belief, AM stations in the United States are not
required to roll off audio above 5 kHz. For many years, no audio filtering
was required, at all! In the early 1990s, however, a set of frequency
response specs called NRSC-1 was adopted. Some type of standard was needed
because, for years, many AM stations had been boosting the treble to
compensate for the poor treble response of many tuners. This increase in
treble had the side effect of causing more adjacent channel interference,
which led radio manufacturers to further narrow the bandwidths of their
tuners. The NRSC-1 spec specified a standard treble boost (pre-emphasis)
curve. In addition to the treble boost, NRSC-1 required a sharp roloff above
10KHz. A related spec, NRSC-2 defined the amount of permissible emissions on
nearby channels. The NRSC-2 requirements can be found in CFR 47, part 73.44.
According to NRSC-2, any emissions 10.2KHz to 20KHz from the carrier
frequency must be at least 25 dB below the carrier. Emissions 20KHz to
30KHz from the carrier must be at least 35 dB below the carrier, and
so on.
9. 5KHz as you can see here.
20KHz wide filter
Of course you could just replace the old ceramic filter with one
of wider bandwidth. These tend to be hard to come by.
Someone gave me a defective but free Sony SS receiver STR-D2020 that
had a blown Sanyo driver module. Replaced that (added to it a bigger heat
sink as it ran kinda hot), and now that it works,
also modified the AM tuner section (changed the IF filter to one that
passes 20KHz to deliver all 10KHz of audio AM stations broadcast, like above). But
I also had to get some tube sound into this box, so I installed a
cathode follower between the function switch chips and the audio
circuit that drives the "soround sound" circuits and regular SS power
amp. This is essentially a case of "tube preamp feeding a SS power
amp". It sounds good. Simulation says I get 2nd harmonic about 45dB
down on a 2Vp-p audio signal, and almost 70dB down on the 3rd. Less
higher up. Well, I like small doses of 2nd H...
And here is a solid state table radio (a Panasonic AM/FM model RE-6280)
that I added a tube audio output to. The tube is a subminiature 5639 "video"
output pentode, not a beam power pentode. Normally this tube isn't all that linear in pentode mode,
so triode mode improves it. But oddly enough, inserting a 5K screen resistor
into the triode wired mode makes it more linear yet. AFAICT, this is just
special to this kind of tube. But this might be applicable to other
oddball tubes otherwise not usable for audio.
