Precision electronic circuits need good quality power. Digital electronics usually combine a voltage regulator module (VRM) with a network of capacitors. The same structure appears in many other types of electronic products.
No matter what the application, all voltage regulator circuits involve the principle of feedback. In any circuit feedback must be carefully controlled because, by its very nature, feedback invites the risk of self-oscillation.
Before diving into a full-fledged analysis of feedback in a VRM application, I shall present a simple example. This circuit example comes from an old Fender Rhodes piano. It has a particularly simple structure that exhibits beautifully the concepts of feedback, conditional stability, and several other interesting problems.
You may already know that the Fender Rhodes piano introduced in 1969 epitomizes the electronic jazz-fusion sound of the pre-synthesizer era. Hear it on Chick Corea's album "Light as a Feather", or Miles Davis' "Bitches Brew". See it in the 1980 film, "The Blues Brothers", where Ray Charles plays "Shake a Tail Feather" in the music store scene. Electronic keyboard instruments made today still strive to achieve that unique, classic "Rhodes" sound.
If you are a Fender Rhodes fan and just can't wait to hear that sound again, here's a live recording of the 'Rhodes, as played by Herbie Hancock.
I love music, but digital electronics in my real passion. I'm looking forward to teaching my first ever Canadian public seminars in Ottawa September 17-20. Then I'm in Dallas October 1-4.
VRM Stability — Part I: Feedback
The instrument on my bench does not sound like Herbie Hancock's piano.
I am looking at a 1972 Fender Rhodes Mark I electronic "suitcase" piano. It belongs to my good friend Chris "Breathe" Frue, a talented jazz multi-instrumentalist. Breathe hovers over the instrument with a somber tone and long face. He has just pulled it from his attic after decades of disuse.
Switching on the power, Breathe grimaces as the piano emits a sound no longer reminiscent of the cool, jazzy, musical marvel of his youth. It buzzes like a swarm of angry bees circulating up from the ground in a whirl of vengeance (those who have heard that sound don't soon forget it). Above the buzzing hoard floats a high, piercing, teapot-squeal, undulating wildly. None of the keys work.
"Breathe", I explain, reaching for a screwdriver, "things may not be as bad as they appear. Self-oscillation requires two things: an amplifier, and a path for feedback. Any sound at all tells us the speakers are OK, probably also much of the power system and parts of the audio amplifier chain. A solution to your problem might only require identifying the source of feedback and cutting it off."
Breathe brightens as he begins to realize that I am not going to merely blow the dust out of the case and start checking transistors. He has never witnessed a methodology for debugging electronics.
At my instruction, Breathe opens up the cabinetry and begins laying the circuits out on my bench.
The Fender Rhodes "suitcase" model comes in two pieces (Figure 1). The keyboard portion rests on the top. It holds the electro-mechanical apparatus and pre-amplifier circuits. The base unit below holds the main connection panel, power amplifier and speakers.
The block diagram comprises five main pieces (Figure 2). First, a set of electronic pickups, one for each of the 88 keys, wires to a common jack (J1). That leads to a pre-amplifier embedded within the keyboard unit (only one of two stereo channels is shown).
The pre-amplifier connects through a shielded cable about 3 feet long leading from the keyboard unit to jack J2 on the main panel, located within the base unit. A second shielded cable hidden entirely within the base unit passes the low-level audio signal through intermediate jack (J3) to a power amplifier. The power amplifier returns the amplified signal through a cable (J4) back to the main panel, from which the signal feeds the speaker (J5). Power connections to the pre-amplifier and the power amplifier are not shown.
Breathe has by now completely extracted the electronics from the cabinetry. All the parts lie exposed on my bench, plugged together in their original configuration. We power on. The system replicates its self-oscillation problem. I welcome this development, as I despise time wasted trying to replicate flaky or erratic problems.
"Begin by isolating the problem," I suggest. "Unplug pieces of the system one by one to see what changes."
Breathe disconnects jack J1, making no difference in the problem. Upon removing J2, the low buzz evaporates. The teapot stays with us, but changes in pitch, warbling. "That's it," exclaims Breathe, jotting down notes on our block diagram, "the buzz must be coming from the pre-amplifier."
"A reasonable guess, but it could also be a problem further downstream." I point to the power amp. "Oscillation problems often depend on the loading of each stage. Unplugging J2 changes the loading at the input of the power amplifier, which could be the real culprit. For now, we can't necessarily tell what causes the buzz, only that you found one thing that stops it. I'm more excited with the fact that the squeal continues. Let's forget the buzz for now and find out what, in the system to the right of J2, causes that squeal."
Inherent to the block diagram, the way I drew it, is a big loop of cables, going from J3 into the power amp, and from there back to the main panel at J4. Even a tiny amount of feedback between these input and output signals could cause the squealing problem.
"It used to work," says Breathe, "twenty years ago. What changed? Why do we get feedback now that wasn't present in the original design?"
"Corrosion." I pull out the connector at J5 and the squeal stops. Of course we stop hearing it, because I just unplugged the speaker. An oscilloscope probe at J4 confirms, however, that the squeal has indeed stopped. This step confirms that our feedback problem involves, somehow, current flowing through the speaker. My test did not change the voltage coming out of the power amplifier, only the current.
"Breathe, let's think about where the speaker current flows in this system." I trace the signal path from the power amp through its cable to J4 and on to the speaker (red arrows in Figure 3). Nothing surprises Breathe about that path. Now I draw another path showing returning signal current coming back from the speaker to the power amplifier. Here lies the mystery of the squeal.
The current returning from the speaker (green arrows) passes along the black wire from the speaker to connector J5. There it shunts to the main panel chassis. From the main panel chassis the returning current must complete the last part of its journey, finding a path somehow from the main panel chassis back to the power amplifier chassis.
Figure 3 represents the last portion of the current pathway with fictional impedance Z1. In actual fact, at audio frequencies the current divides with portions flowing over the ground connections of cables connected to both J3 and J4. The details aren't important—what matters is that this last portion of the path has some finite impedance Z1, so for simplicity I represent it as a single component.
Current traversing Z1 induces a voltage across it. That voltage, V[noise], appears between the main panel chassis and the power amplifier chassis. The magnitude of V[noise] equals the speaker current, I[speaker], times the magnitude of impedance Z1. Now, here's the important part of my discussion: the power amplifier input responds to voltage V[noise].
To see why, short the signal at J2 to its local ground, transmitting no signal beyond that point. The power amplifier input stage now receives the difference between (a) its input pin, connected directly through J2 to the main panel chassis, and (b) its own chassis, which lies at a distinctly different voltage than the main panel due to the returning speaker currents flowing through Z1. The amplifier therefore sees, as its input, a voltage V[noise] proportional to its own output current. When a system sees an input signal that varies in proportion to its own output we call that effect feedback.
A small amount of feedback makes no sensible effect on the power amplifier circuit, however, corrosion present in the ground legs of connectors J3 and J5 may increase the overall resistance of Z1, magnifying the noise voltage V[noise]. If the feedback grows sufficiently large, it can make the power amplifier self-oscillate.
With a feedback theory foremost in my thoughts, I suggest to Breathe that, "Lowering the resistance of the inter-chassis ground connection might help. That would reduce the inter-chassis voltage, possibly suppressing self-oscillation." I reconnect the speaker. The squeal returns as expected. Then I prepare a short piece of bare wire, #10 AWG. With that wire, I short the main panel chassis directly to the chassis of the power amplifier. "See, the squeal stops. Better grounding, less feedback."
Working towards a possible fix, Breathe and I disconnect the cables at J3 and J4 and also at their opposite ends. We clean all the connections. The ground shells on the RCA plugs are covered with a white powder. They appear to be un-plated tin. These get a very light sanding with #600 grit sandpaper. Everything gets scrubbed with CRC Electronic Cleaner, applied with Q-tips. We reconnect, and the system from J2 onwards powers up perfectly with no squeal. Audio signals applied to J2 come through the power amplifier beautifully with no distortion. Squeal fixed.
We are only halfway done with our restoration work, and Breathe has already learned a lot about feedback, amplifier stability, and the importance of good grounding. I will apply those concepts to the study of power system stability in the next issue, where we uncover the genesis of the low-frequency buzz of death—a sound you'd like to hear only in a horror film…
Dr. Howard Johnson
Points to Remember (Summary)
Input current flows into the power amplifier through the jack J3. On its return journey back to the pre-amplifier, the input current traverses impedance Z1.
The output current from the power amplifier flows through jacks J4 and J5 to the speaker. On its return journey back to the power amplifier the output current also traverses impedance Z1.
Impedance Z1 therefore constitutes a common impedance shared by both the loop of input current and the loop of output current associated with the power amplifier (Figure 4).
Common impedance Z1 creates crosstalk (feedback) between the input and output circuits. That feedback occurs in proportion to the magnitude of Z1.
If impedance Z1 grows over time because of corrosion, then feedback grows over time also.
Any amplifying system will oscillate if supplied with sufficient feedback.
Application to Digital Electronics
I encounter grounding problems in high-speed digital circuits all the time. One usually identifies such problems as "ground bounce", or using the more modern term, "simultaneously switching output noise" (SSO). Whatever the name, the concept remains the same: when the inputs and outputs of any circuit share a common current path, the impedance of that path must be kept suitably low.
general, any single-ended system that transmits power in a loop or mesh structure requires exceptionally good grounding. The reference terminals of all the transmitters and receivers in the system must be connected together. They must be connected by impedances so small that the voltages present across those impedances, caused by all the returning signal currents, cannot sensibly affect the circuit.
Grounding becomes especially complicated in digital circuits. An audio circuit only needs suitably low resistance ground connections. A high-speed circuit also needs low inductance connections.