As of a few weeks ago, my monitor (an HP A4576A, a rebranded Sony P1200 or something like that) started tweaking out. The horizontal width would go dip down, and the picture became dark:
Over the following weeks, it got worse and worse. The symptom was intermittent: a binary state of "everything's normal, captian" or bad sweep. A "technical tap" would usually correct it, but the frequency of occurance was rising, and the necessary..."correction", larger. Most recently, it moved into a second failure state, a complete black picture. So, I delved in and started looking for problems.
...A CRT, you say? Yes: I prefer CRTs for their bright, correct colors, high contrast ratio, high resolution (with the market saturation of HD TVs today, try finding a cheap LCD over 1080 pixels high!), wide viewing angle, and rapid refresh rate. Many of these issues have been solved with modern LCDs, but three facts remain: 1. they're expensive, 2. CRTs of this type still remain competitive (after-market prices range from $100-500!), and 3. I don't have a spare—I've been using this monitor continuously for about a decade, and it shows no [other] signs of slowing down! So, why not fix it in my spare time?
Thus begins the teardown:
The construction of these is wonderful. Anyone studying industrial design should open one up and inspect it meticulously. After all these years (Sony introduced the Trinitron back in 1968, and it's been evolving ever since in both size and resolution), these designs have been polished to a fine gloss, the peak of their technology.
The general plan is: the front bezel, CRT, shroud and pedestal are the base unit. Probably it's assembled in the upright position (not face-down as I've done here), around the massive (∼20 kilo!) CRT. The three main boards are assembled on the left, rear and right sides, with accessory boards elsewhere. The left side carries the main deflection board, rear contains the video processing and cathode drivers, and the right side carries the power supply. On the inside floor, there is the magnetometer board (which unfortunately I didn't photograph). There's probably a PCB for the front panel buttons and light (I didn't dig that deep), and there's a small PCB at the back to handle the DVI connector (which I didn't photograph, because who cares). (The VGA connector isn't PCB mounted, it's actually a bulkhead type.)
Already separated, in this photo: the power supply unit. This gives a view into the main chassis; the corner of the magnetometer board is just visible in the shadow. In this chassis, the side modules are held on with three screws, plus slotted catches and alignment bumps—an efficient design, easy to assemble and service.
This is the power supply board. It mounts via plastic snaps to its frame (no screws or ground connections, except for the one wire which connects to a 1/4" QC terminal on the base chasis—a snap-retaining connector, for safety!). Construction is lowest of the low: single sided phenolic, wave soldered! That said, there's evidence of automated test: there are plenty of random pads, round, no hole, soldered over (of course), with a very tiny yet conspicuously sharp pinprick in the middle. Yes, despite the cheap construction, they tolerated no production failures, and did at least some level of bed-of-nails jig testing.
...And yet, for being cheap construction, they did very well. It's hard to tell at a glance, so give the marked-up version a look:
As best I can tell, the layout is this: Power Input (universal 90-250VAC), filtering (some big film caps, 1μF or so, and a reasonable size split-bobbin type common mode choke), then some general AC loads. Everything comes from the power switch, so when the switch is off (it's a satisfyingly clunky push-on/push-off type, though the big button on it feels slightly sloppy), everything is off, no vampire loads. The degaussing coil is powered by a PTC resistor (black box) and relay (other black box), and main power is supplied from another relay (light blue box). Standby, of course, is always on (while the switch is on).
Standby switching controller. Rather than a single transistor blocking oscillator (commonly used for standby in AT/ATX supplies for many years), they went with an integrated IC: the Panasonic MIP0223SY. It's basically a TOPSwitch or whatever. Three terminals: power/feedback, SW and GND. I didn't look in detail, but I would very reasonably guess the circuit is almost straight out of the appnote: these devices are typically shown accompanied by TVS clamps, and the P6KE is visible in front of the device. Besides the standby voltage output, there should be a line-side auxiliary winding, providing self-power for the switcher (maybe 9-12V). Likely, this is also used to supply the PFC controller.
MC33262 Power Factor Controller, and a Shindengen 2SK2195 MOSFET (500V 15A 0.35Ω). The current sense resistors are two 0.33Ω power resistors in parallel, behind the heatsink (visible in the top view). The main inductor is TDK made, some weird-ass bobbin (though not nearly as weird as some of Sony's earlier works). It's wound with multi-stranded wire, technically not Litz (it's not braided), but definitely for the same purpose (it's probably around 50 strands worth). This, in combination with a ferrite (as opposed to powdered iron) core, will give excellent efficiency even at high ripple current (the MC33262 is a boundary control mode chip, meaning current returns to zero on every switching cycle).
Closer shot of the transistor. Oh, and in this picture among others, you can tell it's old equipment—dusty. Although, given its age, it's remarkably clean inside, even around the high voltage—possibly another testament to its design. For the amount of heat (and therefore convection) this thing throws off, I'm not sure how much you can really do about dust control, by design. But whatever they did, it seems to accumulate less dust than most (fanless) things.
After the PFC controller, power comes over to the main inverter. Of course, being Sony, they didn't throw in some cookie-cutter TL494 forward converter or something stupid like that, no. They brought out one of their resonant converters, commonly seen in Trinitrons forever. Obviously, I don't know what their design culture is like, but they must have a special soft spot for circuits like this. Anyway, what seems to be going on here is a "lite" version of some of their more... involved designs:
Pictured is an MX0841. The Internet doesn't seem to know what this is, but fortunately, being top shelf Japanese equipment, they were courteous enough to mark the pins on the silkscreen. They go B1, E1, (unmarked × 2), E2, B2. The unmarked pins clearly complete the pattern as C1 and C2. E1 goes to supply common, C1 connects to E2, and C2 connects to supply positive: a two transistor half bridge circuit. Apparently, it's a freaking dual BJT. Now—when and why they ever decided to fabricate or at least package their own freaking dual power transistor, I have NO idea. But there it is. Very likely, the ratings are similar to an MJE18008, or take-your-pick of popular-at-the-time Japanese switching transistors (2SC2625 saw many years of use, but a vast range of type numbers, ratings and packages have been used). Note that diodes are not used on the board, so these transistors include antiparallel diodes—don't forget to use them if your replacement does not have them internally (which most switching transistors do not).
This is a rough representation of the circuit used. It's similar to the self-oscillating resonant circuit used in CFLs, for example, but the feedback transformer is saturable via a control winding. The effect is presumably this: in normal operation, the feedback / control transformer stays out of saturation, and the oscillator self-commutates (that is, current rises and falls in a sinusoidal fashion, then when current crosses through zero, the transistors swap action and the cycle repeats inverse). But when the control transformer is biased, it saturates sooner, causing the frequency to rise. And, because the load is resonant (maximum power at one frequency), output power can be controlled.
How does that work? First of all, the circuit is a current-feedback half bridge inverter. Essentially the volts-to-amps complement of a Royer oscillator (which uses voltage-controlled switches to commutate a current back and forth): this circuit uses current-controlled switches to commutate a voltage back and forth. If load current flows through the tap of the high-side winding (emphasized with bold lines), this induces positive current in the same winding (providing base current as positive feedback), and opposite voltage to the low side (which remains biased off). When load current reverses, the high side remains biased off and the low side is driven with positive feedback. Important to note: this is actually a bistable circuit—some other mechanism is required to cause it to oscillate.
There are three ways to make this circuit oscillate:
The circuit in question is resonant (as evidenced by the relatively small coupling capacitor), so we can make some other inferences, too. Since the load side doesn't have a filter inductor, there should (must?) be series inductance on the primary side. I don't see another core on the bobbin, or a weird arrangement of windings on the core, but there's enough space in the bobbin to put in stuff like that, either as explicit inductors, or as leakage inductance between windings. It may be air-core inductance, which won't amount to much, so the resonant frequency under short-circuit or startup conditions may be surprisingly high. That's a little troubling, so I wonder a little, what exactly they did.
Now, we can see how control works. Normally, the control transformer stays out of saturation (it might be dimensioned for a ∼50% overhead, I would guess, so that tolerances and temperature variation won't cause it to saturate under any zero-bias condition), and the oscillator operates in the resonant mode. As control bias rises, the effective Bsat of the core drops. At first, nothing happens (the overhead is used up), but as current continues to rise, saturation starts sooner and sooner, and frequency rises. The tricky part happens when frequency is raised to the point where it no longer appears to resonate (the current is a triangular waveform, purely inductive): this far from resonance, the load current is much lower; if it's too low, the control transformer won't even do its job, and oscillation will quench. There can also be spooky pulse patterns and unfortunate resonant frequencies popping up—major downsides of a simplified circuit. Possibly, by selecting values correctly, they were also able to minimize these side effects as well.
It's worth noting that saturation isn't completely for free. There was an unusually well produced overunity scam back in 2007 or so (this looks like the thing, but the exact video I recall was a surprisingly well produced press release). At the time, I figured, off the cuff, that—of course there's power flowing somewhere, they're just missing it—and the question of where is often an intriguing one to follow in problems like these.
Naturally, if you have a situation where you're biasing a core into saturation, and sensing that elsewhere, even if it's with an orthogonal winding (as in the schematic above: the windings are at right angles, so each one sees net zero flux from the other, assuming the core remains linear!), the sheer act of sensing necessarily requires that you apply more than zero flux to that winding. Which will cause one leg to push closer to saturation, and another to push away from it. The leg that saturates decouples the windings, increasing coupling through the remaining three legs. In the case of the motor, they were willfully neglecting the induction of the magnet passing by the now-unbalanced control winding, and avoiding its energy expense in the calculation; in the case of this control transformer, the consequence is that some of the load voltage ends up coupled into the control winding.
The consequence of THAT, in turn, is nothing special: it means that, if you don't want to interfere with the power circuit operation, you should avoid loading the control winding with a constant voltage. In practice, a moderate resistance is sufficient, or you can drive it with a constant current source (like a BJT collector), which is probably what's done here. But it does mean you want to be careful about connecting it just willy-nilly; you wouldn't want to drive the control winding with a voltage.
Anyway, that covers the power supply. Moving on, this is the video amplifier / processor board: the "A" board. It's actually two, but the CRT socket board is clearly part of the assembly, to the extent that it's even cut form the same PCB at assembly time (the outlines match perfectly, minus the router kerf, and the breakaway tabs are clearly visible on both).
Power enters via the rainbow wires on the bottom connector, and the others bring in, I think, control signals from the MCU on the deflection board. The black and white wires are two sets of three (R, G, B) coax cables, leading from the rear panel (the two signal sources, VGA and DVI, are selected in the OSD menu). The video must be impedance controlled, well shielded and grounded, coaxial all the way—with pixel clocks over 200MHz, analog bandwidth must be over 100MHz to maintain good, sharp pixels, and the dispersion (droop, attenuation) needs to be in the <1dB range for accurate (8 bits per channel) levels.
The main amplifier here is a monolithic (or maybe it's hybrid) Hitachi HA4111. Not much data on it, but I see hints that it's a 120MHz bandwidth, three channel, high voltage amplifier, just what's needed. Rise/fall time of a couple nanoseconds, and a pretty wide voltage range. Being a vacuum tube, the CRT needs fair voltages to do its job, around 80V driving the cathodes. These amplifiers are pretty specialized: there's nothing else that needs high bandwidth in as much voltage. So, for the few odd times that you find you need this kind of range, you're kind of stuck on what you can find to do it.
Today, the available products break down thusly: RF amplifier transistors are all lower voltage (24 to 42V supply, meaning a breakdown voltage in the 50-120V range), because matching circuits are used to transform and combine them into whatever system impedance or power level is required. Switching transistors (MOSFET or otherwise) are all much too slow (too much gate capacitance). There are some applications for higher voltage RF transistors, usually something like high efficiency 13.56MHz generators for industrial use (ranging from lighting to semiconductor manufacture), but they're hard to find.
This is the CRT base, as it fits into the socket board. The plastic base has several wings, which provide creepage clearance for the high voltage (G2 and focus) pins; these features are repeated in the socket.
The ribbon cable from the "U"-shaped board is labeled, from left to right: GND, Heater, GND, KR, GND, KG, GND, KB. The cathode drive is all but done at this point, but the ribbon still alternates between grounds to keep signal quality good. On the neck board, 22 ohm resistors provide some damping and ESD protection, while neon ligts (most likely, a GDT more refined than a generic neon light) provide clamping in case of internal breakdown. (CRTs have been known to arc over internally from time to time—your equipment isn't going to last long if you aren't expecting operating modes like that.)
It's worth noting, this is way worse than any kind of ESD you see normally: we're talking nearly 30kV, with lots of capacitance behind it (∼nF?) and low resistance—this puts Machine Model ESD to shame! The resistors are beefy (1/2W) carbon composition, and there are protection diodes near the amplifier. This level of protection is a serious challenge around high bandwidth amplifiers, and they followed the correct approach here, by all means.
Needless to say, use caution around all these boards: the video amp uses 80 and 200V, G2 uses 1500V, and the deflection board uses everything inbetween (plus some crazy waveforms). Always check that capacitors have discharged to a zero-energy state, or if you're servicing it live... well, you're better off not doing that...
This is the deflection board. A bad view, because I didn't want to pull it completely out: it has many connectors, plus the high voltage lines. A dangerous board, and also the most important.
Now, wait... what is THAT? Brown goo, it looks like? That can't be good...
It IS brown goo! It looks like that... black "456" thing shit itself something awful, maybe.
After later removing the board for closer inspection, I played with the goo a bit: it has no smell (or taste..), it's thick, gooey but not sticky, a very slow moving liquid at room temperature. It's very soluble in acetone. Possibly, it started seeping maybe five years ago, and it's only now oozed down as far as it has. Which is problematic, because it's seeped into a connector (see the next photo), which made it awkward to pull that one out. Maybe... some goo wedged itself into that connector, springing the contacts, and giving me this intermittent failure?
I desoldered the "456" component. It appears to be a standard ferrite core inductor, wound with litz. Go figure with Sony always having to make their own weird bobbins and cases. Still checks continuity (< 0.3Ω), didn't try measuring further. It looks hollow, like there was supposed to be potting inside. Hmm. I removed the other one ("455") for comparison; it's still continent, and appears to be filled with a soft rubbery material. Perhaps the potting on the one was devulcanized, somehow; maybe a bad formulation, maybe something environmental and autocatalytic, who knows.
After replacing the inductors, snooping around for cold solder joints (and finding none—not that that's a conclusive statement), and cleaning the affected connector (a good soak in acetone cleaned it up nicely), I put it back together. Hey, it works!...damn, it doesn't. Still flaky, though it worked without problems for a few hours anyway. Damn, still something in there. But WHAT?
More views of the board. The rectangular TQFP is a Fujitsu F2MC-16LX series, 16 bit MCU with mask ROM, running at 16MHz. Most likely, the main processor (in charge of EEPROM, OSD, service comms, etc.). There may also be a DSP floating around to handle real deflection / video tasks (on this board or the "A" board), or this one may have custom peripherals inside to handle those sorts of things. The chips beside are CXA2043 and -44, no data on them; certainly something deflection or DSP related.
Further over, tucked into the corner: the vertical deflection amplifier. This is kind of an odd package (something like a wide TO-3P with unusually thin heatsink tab, or a HEPTAWATT-style package with a different outline), but it appears to be a pretty standard monolithic deflection amp.
Deflection board underside. (Warning, I took several pictures and stitched them together, so it looks patchy and some parts may be missing or something. Full size 1.47MB.) The most prominent features here are the heavy (high current) traces, the high voltage traces (lots of clearance, silkscreened over!), and the many routes to cover creepage. The heatsink is at the top, with the deflection output centered (right on the white stripe). S-correction takes up the middle (five film capacitors and seven MOSFET switches). Much of the rest is, I believe, pincushion drivers and stuff like that. High voltage goes in the bottom left corner; the FBT has its own dedicated switch (a MOSFET, actually), which is below and to the right of the "HRC" part.
More than a few SMT components reside on this layer, including some SOICs and one TSSOP. Note the wave soldering footprints—extra pads off the ends, which soak up extra solder so the SMT pins don't get bridged. Most of the chip (resistor/capacitor) parts also have thin traces leading away from them. It looks like the wave direction was intended to be bottom to top, though I haven't noticed any markings on any of the boards to go with that (sometimes, boards are marked in silkscreen or copper as to which direction they should be ran.)
So, at this point, I still haven't found the problem (although, if you look closely, it both is, and is not, looking me in the face!). I resigned to doing an open frame test. This is the setup: the deflection board, stripped of its mounting bracket, rests on the bottom inside chassis. It didn't look like there are any important traces along the bottom, so hopefully it won't short anything there. The heatsink is wedged between the front shroud and the CRT itself, giving some stability—but if the board slides rearward a bit, there's not much keeping it from flopping out and zapping things / itself!
So, I proceeded very carefully, poking, prodding, tapping and dragging at the circuit with a wooden skewer. I had determined that it was most sensitive in the area around L504 (the one that de-potted itself), but that was still very inconclusive: flexing the board in that area could still be cold solder joings anywhere from the heatsink to the FBT.
Then, one pass found an immediate result: merely brushing over a chip resistor, R501, caused the deflection to return or drop out! (Blurry picture: in my defense, it was single handed, somehow holding the camera and flashlight in one hand and pressing the button, and pointing the stick with the other.) It wasn't cold solder joints, it wasn't loose connectors, it wasn't flaky transformers or inductors or blown capacitors or transistors, it was a freaking chip resistor! Go figure!
No idea what happened to it, exactly: it doesn't look damaged, and after desoldering, it still measured 4.71kΩ. I replaced it with a 1/4W carbon film resistor, leads bent to fit the pads—just in case, if it was due to peak loading, a meatier resistor will be less likely to fail again.
Button it all back up and, guess what, it works a treat! Probably good for another decade.