Choosing the audio capacitors

The motor run caps were about keeping the transport alive. The next round is the audio boards — PA-285, PA-286 and PA-287, the schematics for which this site now carries — and it is a different kind of shopping list. Across the three boards there are just over sixty electrolytic positions, and about a third of them are coupling capacitors: parts whose entire job is to pass the music and block the DC.

Why leakage is the spec that matters

A coupling cap sits between two circuit nodes at different DC voltages and is supposed to let none of that DC through. The current that sneaks through anyway is the leakage current, and on this machine some of those positions face the heads: the replay-head input to the EQ amplifier, the sel-rep pickup from the record head, and the record amplifier's output coupling into the record head itself. DC through a replay head slowly magnetises it, and a magnetised head lays hiss onto every tape it plays and shaves a little more off the top end with each pass — including, eventually, tapes you care about. (Distortion and pops are the record- and erase-head versions of the same disease.) Elsewhere the leakage lands on potentiometers, where it makes them scratchy, or on the next stage's bias point, where it shifts the operating point the designer chose.

The factory's answer in 1977 was tantalum capacitors in the head-facing positions. Tantalums have excellent leakage when new — and a famous retirement plan: they fail leaky, and then they fail short. Every one of them is going.

The industry writes leakage as a formula on the datasheet. Standard aluminium electrolytics are specified at I ≤ 0.01·C·V — capacitance times rated voltage, in microamps. The low-leakage grade is I ≤ 0.002·C·V, five times tighter. For a 10 µF/50 V part that is the difference between 5 µA and 1 µA allowed through. Small numbers — and in fairness, the steady microamps are the lesser half of the threat: the practitioner rule of thumb puts the danger line well above them. The greater half is what a capacitor does in its final hour. A failing part can release current through the head in one direction, once — and one transient is all magnetisation takes. That is the tantalum retirement plan, below, and the real reason the head-facing positions go first.

The discontinuation problem

Here the hunt turned into a small lesson in reading datasheets rather than series names.

• Nichicon UKL — the low-leakage series, 0.002CV, the natural choice — carries a banner on its current datasheet: "Products which are scheduled to be discontinued. Not recommended for new designs." The best coupling capacitor you can buy for this machine is on its way out. What distributors hold is what there will ever be.
• Würth WCAP-ATLL — I assumed from the name that LL meant low leakage. The datasheet says otherwise: it is a Long Life series, and its leakage values work out to the standard 0.01CV with a 3 µA floor.
• Würth WCAP-ATUL — marketed as "Low Leakage & Long Life", with coupling listed among its applications. The part datasheet I sampled specifies 0.01CV. The marketing and the spec table disagree, and the spec table wins.
• Vishay 013 RLC — the quiet survivor. The datasheet specifies IL2 ≤ 0.002·C·V + 1 µA after two minutes, it is in active production, and my 47 µF/50 V parts are rated 5.7 µA. The catch: the series runs 33 µF to 470 µF. Nothing smaller.

That last catch looked fatal, because the most common value on these boards — around forty-four positions, including every former tantalum — is 10 µF. The first version of this entry declared the guaranteed-spec era over at that value: the UKL listing I checked read obsolete with zero stock, and nothing else I searched specified 0.002CV at 10 µF.

Both halves of that conclusion were wrong, and the mistake is worth logging: I had been searching by the series names I already knew. A systematic second sweep — reading spec tables instead of typing in names — found Rubycon TWL: in active production, leakage specified at 0.002·C·V or 0.4 µA, whichever is greater (datasheet), with 10 µF stocked at 16 V and 25 V in the thousands. And the UKL door isn't shut either: the 10 µF/100 V UKL sits at the distributors in quantity, marked not-for-new-designs — last-time stock, but buyable today. At 10 µF the guaranteed-spec era is alive; it just doesn't answer to the names I was searching for.

The plan

• 10 µF couplings (every former tantalum among them) — Nichicon UKL 10 µF/100 V, sixty bought deep from the not-for-new-designs stock: the guaranteed low-leakage grade at exactly the voltage headroom the next section argues for. Twenty Rubycon TWL 10 µF/25 V ride along as the active-production backstop. Each part still gets its leakage measured on the bench before it goes in.
• 1 µF couplings — Nichicon UKL 1 µF/100 V. Seven positions; I bought seventy-five, because at these prices depth of stock beats a second EOL hunt later.
• 33 µF line-output couplings — Rubycon TWL 33 µF/25 V: the same guaranteed series at the value that couples the line amplifier to the XLR jacks. (The Vishay 013 RLC covers 33 µF on paper, but it is active-with-zero-stock at the distributors right now — eleven-week lead times are for other people's projects.)
• 47 µF positions — Vishay 013 RLC 47 µF/50 V, the active low-leakage series.
• 100 µF and 220 µF — UKL again, 63 V and 35 V.
• Emitter bypasses and rails — leakage doesn't matter there; ESR and endurance do. Panasonic FR and FS, low-ESR and 5000–10000 hours at 105 °C, take those positions.

On voltage ratings, again

Everything on the list is rated well above the 1977 originals — 50 V and 100 V parts into 16 V and 25 V positions. As the motor caps already taught: the voltage rating is a maximum, not a set value, and the headroom is free reliability.

For leakage it is better than free. The datasheet formula is specified at rated voltage — 0.01CV for a 10 µF/100 V part means 10 µA with a hundred volts across it. In an audio coupling position the cap sees a few volts. Leakage in an aluminium electrolytic falls steeply below rated voltage — the manufacturer curves (fig. 7–9 in the Vishay sheet are typical) show a part at a tenth of its rating leaking a few percent of its rated-voltage figure. So a healthy standard-grade 100 V part at 5 V bias will, in practice, likely leak less than a low-leakage 50 V part's printed limit. The printed limit is a guarantee; the curve is typical behaviour. After the correction above, every coupling value on the list can carry the guaranteed 0.002CV grade — so they all do.

The bench check stays anyway, promoted from workaround to selection: a guarantee is a ceiling printed for a whole production run, and the measurement is the number for the one part going into the one position. Every coupling cap gets it before soldering: the candidate in series with a DMM on its microamp range and a 1 kΩ resistor to tame the inrush, across a bench supply at the working voltage; the datasheet's two minutes for the dielectric to polarise, then read, then discharge through the same resistor. A good part settles into the fractions of a microamp; a bad one announces itself immediately. Two minutes per part, and the spec argument is over — replaced by a measurement of the exact capacitor going into the exact position.

Run a 100-volt part at 5 volts and its leakage all but disappears.

leakage curve

Applied voltage5 V

At 5 V this 100 V part typically leaks 0.01 µA — under the 1 µA printed limit the old low-leakage grade guaranteed. The printed limit is a promise at full voltage; the curve is what the part actually does at yours.

The numbers

Datasheet limits are specified at rated voltage: standard grade ≤ 0.01·C·V µA, low-leakage grade ≤ 0.002·C·V µA. Below rated voltage leakage falls steeply (manufacturer curves: ~3% of the rated-voltage figure at 20% of rated voltage). Curve shown: typical 10 µF/100 V part; dashed line: the 1 µA printed limit of the old low-leakage grade at 10 µF/50 V.

The pattern lesson

Same as the 5.5 µF hunt, one shelf over: the series name is advertising, the spec table is the contract. "LL" in a part number told me nothing; a banner on page one of a datasheet told me to buy the parts now or never; and the one series that actually guarantees the number I care about doesn't put it in the name at all. And one more for the shelf: a distributor search is a snapshot taken through the keyhole of the part numbers you already know. The second sweep found, in active production and in stock, the exact thing the first sweep swore no longer existed.