Tuning the X68000 Audio Output Filters With a Spectrum Analyzer

The Sharp X68000 has always punched above its weight when it comes to audio, blending a custom Yamaha YM2151 sound chip with an OKI MSM6258V for sampled playback and feeding the whole thing through a discrete analogue stage before it ever reaches a pair of speakers. Anyone who has hooked one up to a decent amplifier knows that the stock output can sound slightly veiled in the upper midrange, and the culprit is rarely the chips themselves. It is almost always the passive filter network sitting between the DAC and the RCA sockets. Realigning that stage takes more than a soldering iron and a hopeful ear; it calls for proper signal analysis and a measured approach.

A spectrum analyser turns subjective hunches into hard numbers. It sweeps a known signal through the audio chain and plots amplitude against frequency, exposing roll-off points, resonant peaks and unwanted harmonics that would otherwise hide under music. For Australian hobbyists used to pulling salvaged parts from a mate's shed or driving down to Jaycar for a bag of polyester caps, this kind of rig used to live only in university labs. Affordable USB analysers and secondhand HP units from the 1990s have changed that, putting real measurement capability on a kitchen table in Melbourne or a spare room in Brisbane.

Working on a machine like the X68000 demands patience and the right vocabulary. The community around these computers is deeply technical, trading schematics and meeting at small gatherings, sometimes in the back room of a suburban RSL club, sometimes at a dedicated event in Sydney. Sharing measurement results across these groups keeps the platform moving forward, because every documented tweak saves the next restorer a few weekends of trial and error.

The reward for sitting down with the analyser is an audio path that finally sounds the way the engineers at Sharp originally intended. With the right test signal, a clear reference trace and a handful of replacement parts, the filters can be coaxed back into spec or even pushed slightly beyond it for cleaner highs and a tighter low end.

Why the Audio Filter Stage Matters

The X68000 routes both the FM synthesis output and the ADPCM samples through a small analogue network before they reach the rear panel jacks. On most revisions this network combines a passive low-pass section for smoothing the PWM-like output of the YM2151 and a simple AC-coupling stage to remove DC offset. The values chosen by Sharp were a compromise: tight enough to keep clock noise out of the audible band, gentle enough to avoid dulling the bright character of the FM patches. Three decades of electrolytic capacitors drying out, tantalum parts drifting and trimmer resistors oxidising have shifted that balance noticeably.

Once the filter drifts, the symptoms show up in different ways depending on which way the components have moved. A drying coupling cap rolls off the bottom end, leaving percussion thin and vocals recessed. A leaky filter cap does the opposite, dumping low frequency noise straight into the signal path and amplifying mains hum, particularly noticeable on Australian 50 Hz mains. Either way, the music loses the crisp, slightly aggressive edge that makes games like Akumajou Dracula and Gradius II feel so immediate on the original hardware.

Restorers often start by replacing the obvious culprits, the electrolytics, and hope the problem goes away. Sometimes it does. More often the board comes back sounding marginally better but still not right, because the replacement caps are close to but not exactly the original values. Even a ten percent tolerance swing on a single resistor can shift the corner frequency by a noticeable amount, and that is before any tolerances stack up across the rest of the network. Knowing which part of the response has shifted, and by how much, is what separates a rough repair from a properly tuned restoration.

The Spectrum Analyser as a Tuning Tool

A spectrum analyser for audio work does not need to be the size of a small fridge. Modern USB units connect to a laptop and run sweeps up to 96 kHz, more than enough to capture everything the X68000 produces and the unwanted harmonics above it. Older benchtop analysers from the 1990s, often sold through Australian surplus dealers and online auctions, still perform brilliantly and bring the satisfaction of proper front-panel controls.

The technique used to tune a filter is straightforward in principle. A pink noise source or a stepped sine generator feeds a known signal into the X68000's audio input path or, more commonly, replaces the upstream stage while the filter under test stays in circuit. The analyser plots the resulting amplitude at each frequency, building up a curve that mirrors the filter's transfer function. Where the curve dips, the filter is attenuating; where it peaks, resonance is building up; where it should be flat but slopes gently away, components have drifted.

Software packages popular in the local retro scene include Room EQ Wizard, ARTA and the older Holm Acoustics tools, all of which can drive a calibrated soundcard or a dedicated measurement interface. The results are repeatable, the curves are easy to save and compare, and the methodology is the same whether the bench is in a Perth garage or a Hobart flat. Crucially, the analyser exposes problems that the ear alone attributes to "the recording" or "the mastering", which is why documenting each session in a project diary becomes so valuable.

Setting Up the Measurement Environment

Before any sweeping begins, the test setup itself needs attention. The X68000 must run from a stable supply, ideally the original PSU rebuilt with fresh caps or a known-good replacement such as the Nereid-X compatible unit, because any rail ripple shows up directly in the measurement. Mains-borne interference at 50 Hz and its harmonics can dominate a sweep if the chassis is not earthed properly, a quirk that catches out Australian hobbyists whose house wiring sometimes lacks a dedicated earth at older switchboard installations.

The output of the analyser connects to the X68000's line-out through a high-impedance probe or a simple buffer, ensuring the test gear does not load the filter. A 10:1 oscilloscope probe works well for probing individual nodes on the PCB, while a line-level input feeds the final stage for end-to-end measurements. Cables matter too; anything beyond two metres starts to roll off the top end and produce misleading curves at 20 kHz.

Acoustic measurements, where the analyser drives a speaker and uses a calibrated microphone to capture what reaches the listener's ear, are tempting but rarely necessary for filter tuning on the X68000. The analogue stage is best characterised electrically, with the speaker and room variables removed from the equation. That keeps the focus on what can actually be fixed with a soldering iron and a fresh strip of polyester capacitors.

Analysing the Stock Filter Response

With a sweep running, the stock X68000 filter network typically reveals a gentle shelf above 8 kHz, with a steeper roll-off kicking in around 15 kHz to suppress the FM chip's switching artefacts. On a healthy board the response stays flat within a couple of decibels across the midrange. If the analyser shows a sloping curve that loses three or four decibels by 5 kHz, the coupling capacitors are usually the cause.

A more subtle problem appears as a small peak somewhere between 12 kHz and 18 kHz, often only a decibel or two high but enough to give cymbals a glassy, almost metallic edge. That peak is usually a resonance in the filter stage, created by the parasitic inductance of an electrolytic interacting with the input impedance of whatever amplifier follows it. Identifying it precisely requires zooming into the analyser display and stepping through narrow frequency bands rather than relying on a single broadband sweep.

Recording the baseline trace before any changes are made is essential. Saving the file with a date and a board revision note, and dropping a short write-up into the project log on x68k.net/diary, means future comparisons stay honest. It also gives the rest of the community a reference point, which is how knowledge of these machines accumulates one careful measurement at a time.

Component Swaps and Recalibration

Once the baseline is locked in, the next step is component replacement. For the coupling stages, modern polypropylene capacitors in the same value as the originals bring the bass response back without introducing the dryness that some older polyester parts develop. A common upgrade in Australian workshops is to substitute a 10 µF / 50 V electrolytic with a 10 µF / 63 V film type, gaining tighter bass and lower distortion at the cost of a slightly larger footprint that sometimes needs creative mounting.

The filter caps themselves often respond well to swapping in tight-tolerance parts, particularly in the second-order stages where two capacitors work together. Moving from a 20 percent electrolytic to a 5 percent film or tantalum unit pulls the corner frequency closer to the design value and smooths out the slope of the roll-off. Where a trimmer exists on the board, the analyser makes short work of setting it: adjust the trimmer while watching the curve until the peak or dip disappears, then lock it in with a dab of nail polish.

It is tempting to keep tweaking, but a disciplined approach saves time. Change one component at a time, re-run the sweep, log the result, and only then decide whether the next swap is worth it. That way, if something goes wrong, the cause is obvious and the board can be returned to a known state. The community has learned this through years of late-night troubleshooting, often over a servo of flat white in a suburban kitchen, and the wisdom now travels freely through mailing lists and Discord servers.

Verifying the Tuned Response

With the filter network rebuilt, the final pass is verification. The analyser should now show a response that is flatter through the midrange, with a clean roll-off above the chosen corner frequency and no resonant bumps. Listening tests then confirm what the numbers suggested: snare drums snap, bass lines carry weight without boom, and the high-frequency sheen of FM patches returns without the artificial edge that characterised the drifted board.

It is worth running the sweep again a week later, after the new capacitors have had time to settle and any mechanical stress from soldering has relaxed. Drift in the first forty-eight hours is rare but not unheard of, particularly with certain film types, and catching it early avoids the disappointment of a system that sounds perfect one day and slightly off the next. Saving the final trace alongside photos of the rebuilt section closes the loop and creates a reference that the next restorer can build on.

Tuning the X68000 audio path with a spectrum analyser is one of those jobs that looks intimidating from the outside but becomes very approachable once the first trace appears on screen. The combination of careful measurement, patient component swaps and honest documentation transforms a slightly tired vintage computer into something that genuinely sings, and the techniques learned along the way transfer neatly to any other piece of analogue gear worth saving.

For further reading, build notes and progress photos covering this and other X68000 restoration projects, the latest diary entries walk through capacitor selection, filter theory and a number of before-and-after sweep traces captured on the bench.