Frequency Division & Accuracy

Ask how accurately a neon ring counter divides a frequency and you have asked a question with a surprisingly sharp two-part answer, and getting the two parts straight is the whole point of this volume. The tidy part is that a ring counter is a counting divider: it counts input pulses and emits one output pulse for every N it receives, so the division ratio is the integer N — not N plus a little, not N on a good day, but exactly N — and it stays exactly N for as long as the ring counts reliably. A neon divider therefore adds no frequency error of its own: feed it a source of f_in and its output is f_in/N to the same precision as f_in itself. (The Philips Philicorda organ at the end of this volume turns out to divide the other way — with synchronised neon oscillators rather than counting rings — which makes it a useful contrast.) The untidy part — and the part a reader who has actually run one of these will nod grimly at — is that “for as long as the ring counts reliably” is doing enormous work in that sentence. A cold-cathode ring is only as trustworthy as its tube margins, and when those margins erode two distinct defects appear: reliability faults (a tired stage that drops or doubles a transfer, corrupting the count) and jitter (the transfer instant wandering in time, smearing the output edges). Neither is a change in the ratio; both are ways the exact ratio can be broken or blurred. This volume pulls those threads apart, quantifies each, distinguishes the ring from a fundamentally riskier kind of divider that can land on a wrong ratio, works the PA3FWM mains-to-minute divider chain as the canonical cascade, and then tells the Philicorda story.

6.1 A ring is a counting divider — and counting divides exactly

Recall the mechanism from Vol 4: N neon stages share a common anode, exactly one is lit, and each input pulse extinguishes the lit stage and lights the next one along, so the glow marches around the ring one place per pulse. Tap a fixed stage — say the last one — and that stage lights once every time the glow completes a full lap, which is once every N input pulses. That single fact is the entire theory of neon frequency division: one output event per N input events. There is no averaging, no phase-locked loop settling, no analogue quantity that can be a few percent off. The output is a consequence of counting, and counting to N gives you N — an integer, by construction.

Put it as an equation, defining every term. Let f_in be the input pulse rate in hertz, N the number of ring stages (a positive integer), and f_out the rate at the carry tap. Then

f_out = f_in / N — with N exact.

Contrast this with an analogue divider such as a resonant tank or a regenerative mixer, where the output frequency depends on component values that carry tolerances (±5 % capacitors, temperature-drifting inductors) and therefore inherit those tolerances as a frequency error. The ring has no such term. Its N is set by how many stages you wired, a quantity known to infinite precision because you can count the lamps by eye. If the ring counts at all, it counts to the right number. This is the same reason a digital flip-flop divider (Vol 6.7) is trusted to divide a crystal down to a wristwatch’s one pulse per second: division by counting is division by an integer, and integers do not drift.

So the headline is genuinely reassuring: the neon stage contributes nothing to the frequency accuracy of the divided output. All of the frequency accuracy you get out is frequency accuracy you put in. If you want a pitch stable to a few parts per million, you put a few-parts-per-million source in front of the ring; the ring will hand you that stability divided down, untouched. Everything difficult about neon dividers is therefore not about the ratio. It is about the two ways the counting itself can misbehave, which occupy the next two sections.

6.2 Reliability: the ratio is exact, but a mis-count breaks it

The first way to spoil a perfect ratio is to break the counting. A ring divides by N only if it advances exactly one stage per input pulse — no more, no less. Two failure modes violate that, and both were catalogued the hard way by the builders in our spine sources (Vol 1’s Dekker, Luc Small, and PA3FWM).

A dropout happens when an input pulse fails to move the glow: the pre-biased next stage does not reach its striking voltage Vs, either because the carry pulse was too weak or because that particular tube’s Vs has crept up with age (Vol 7). The glow stays put, the ring has counted one input as zero, and the divided output is now one pulse short over that lap. A double-step is the opposite: a single input advances the glow two stages, because the coupling capacitor C0 was oversized (Vol 4’s warning) or two adjacent tubes’ margins overlapped so that extinguishing one primed two neighbours into striking. Now the ring has counted one input as two, and the output runs one pulse long.

Notice the character of the damage. A dropout or double-step is a ±1 integer error in the count, and in a divider a ±1 count error is not a subtle few-cents detuning — it is a gross, discrete event. In a clock (the PA3FWM machine of §6.5) it is a lost or gained second that then propagates up the whole chain. In a musical octave divider built from counting stages, a ÷2 stage that momentarily divides by 1 or by 4 throws the note a whole octave — an unmistakable glitch, not a wobble. This is the crucial accuracy nuance and worth stating flatly: when a neon divider fails, it fails by an octave (or a whole count), never by a few parts per million. The frequency you get is either dead right or conspicuously, discretely wrong. There is no graceful middle where the pitch simply drifts sharp, because there is no analogue quantity in the count path to drift.

The mitigations are exactly the disciplines the rest of this series preaches, and they all serve reliability, not ratio: bin your tubes so every stage has a comfortable Vs−Vm window (Vols 3, 11); burn them in first so their voltages have stopped moving (Vol 11); run a stiff, regulated supply so the margins do not sag under load (Vols 8, 9); centre the cathode bias at (Vs−Vm)/2 and prefer wide-window switching tubes like the ZA1002 (~60 V window versus an NE-2’s marginal ~12–20 V); and where a weak modern-lamp carry pulse cannot reliably clock the next stage, add a buffer/amplifier lamp on the carry, biased just below its strike voltage so a small pulse tips it over (the PA3FWM fix, §6.5). None of these touch the ratio — the ratio was always exactly N. They keep the ring counting so that the exact ratio is actually delivered.

6.3 Jitter: the ratio is exact, but the edges wander

The second defect is subtler and does not corrupt the count at all. Even a ring that never drops or doubles a transfer does not place its output edges at perfectly even intervals. The instant at which a stage strikes depends on how fast the anode voltage recovers through Ra, on the exact Vs of that tube at that moment, on the temperature, on supply ripple, and on how well-primed the gas is (the photosensitivity and dark-effect of Vol 2). All of those quantities fluctuate slightly pulse to pulse, so the transfer instant wanders in time by some small Δt. That timing uncertainty is jitter, and on the output waveform it shows up as phase noise — the edges are in the right place on average, but each individual edge arrives a little early or a little late.

The distinction that matters for accuracy: jitter is not a frequency error. Averaged over many cycles the count is still exactly N, so the long-term frequency f_out = f_in/N is untouched — over a minute, a ÷3000 chain still delivers exactly one pulse per minute even if every individual edge trembled. What jitter degrades is the short-term purity of the edges: the cycle-to-cycle period is T_out ± Δt. In a clock that is invisible (a wristwatch does not care that its one-per-second edge arrived a microsecond early). In audio it is audible only as a faint noise or “grain” on the tone, not as a pitch error — which is again why, in an organ built from counting dividers, the tuning is set by the oscillators while the dividers contribute, at worst, a little textural roughness. It is helpful to keep the two-axis picture of Figure 6.2 in mind: reliability is whether the count is right (a discrete, catastrophic axis), and jitter is how cleanly each correct edge is placed (a continuous, cosmetic axis). Neither is a ratio error, and that is the recurring theme.

It is worth putting a rough number on the jitter so its scale is concrete. The transfer instant is set by how quickly the anode recovers to Vs after the input pulse, which is an Ra·C charging edge; a small fractional wobble in Vs maps to a proportional wobble in the crossing time. With Dekker’s values (Ra ≈ 82 kΩ, a stray-plus-coupling capacitance of order tens of nanofarads) the recovery time constant is on the order of a millisecond, so a Vs that flickers by even 1 % — entirely plausible for an ageing indicator neon in changing light — translates to a transfer-timing wander of order tens of microseconds. Against a per-stage period near the ~1 kHz ceiling (1 ms) that is a percent-level phase smear on each edge, and against a slow output (a 1-per-second or 1-per-minute clock tick) it is utterly negligible. That is the right intuition: neon jitter is large compared with a crystal oscillator’s but tiny compared with the intervals a neon divider is usually asked to mark, which is why it is a cosmetic defect in practice and never a timekeeping one. A well-primed, wide-window switching tube on a clean rail brings the figure down further; a starved, dark-adapted NE-2 pushes it up and, past some point, tips over from mere jitter into the outright dropouts of §6.2.

Jitter is fought with the same stiff-supply-and-priming hygiene as reliability, plus one extra: keep the anode recovery fast and repeatable. A well-defined Ra·(stage capacitance) recovery, a clean regulated rail with low ripple, and reliable priming (a keep-alive glow, ambient light, or a primed switching tube such as the tritium-primed ZA1002 that strikes in under 1 ms) all shrink Δt. Low-leakage, low-dielectric-absorption coupling capacitors (Vol 10 — polypropylene or C0G, never electrolytic or high-K ceramic) keep the transfer timing from wandering as the caps’ effective value shifts. But the ceiling is physical: cold-cathode deionisation takes on the order of a millisecond, which is both why a neon ring tops out near ~1 kHz per stage (Vol 2) and why its edges will never be as crisp as a silicon flip-flop’s. You divide accurately in ratio but softly in time.

6.4 A true counting divider vs. an injection-locked one

Here is the failure mode a ring counter is gloriously immune to, and understanding why is the deepest part of the accuracy story. There is an entirely different family of dividers — injection-locked or regenerative dividers — that produce a sub-harmonic not by counting but by pulling a free-running oscillator into step with a drive signal. A relaxation oscillator (including a bare neon-lamp relaxation oscillator, Vol 2) that free-runs near f_in/N will, when you inject the drive f_in, lock to the exact sub-harmonic f_in/N as long as f_in/N falls inside its capture (“lock”) band. It looks like a divider and, in its lock band, it is one.

The danger is what happens at the edges of that band. If the drive amplitude sags, temperature shifts the free-running frequency, or the supply moves, an injection-locked divider can unlock and re-lock on a different sub-harmonic — jumping from f_in/N to f_in/(N−1) or f_in/(N+1). It does not glitch once and recover; it settles, quietly and stably, on the wrong ratio, and it can stay there looking perfectly healthy. That is a qualitatively worse failure than anything a ring can do, because it is a wrong frequency that is stable — exactly the thing you cannot catch by ear or by a casual scope glance.

A counting divider cannot do this. The ring has no free-running frequency to be pulled; it is a discrete state machine that emits a carry only after N genuine input events. Its ratio is topologically fixed at N by the number of stages. Starve it, and it does not slide to N−1; it drops or doubles a single transfer (§6.2) and then resumes dividing by N. The distinction is worth memorising: an injection-locked divider can be quietly wrong about the ratio; a counting divider can only be momentarily wrong about the count. For anything where the ratio must be trusted — a clock, an organ’s octaves — you want a counting divider, and the neon ring is one. (The Philicorda organ of §6.6 is, instructively, the other kind: its ÷2 stages are neon relaxation oscillators synchronised to the stage above, not counting rings, and owners report exactly this wrong-ratio failure — a stage dividing by 3, or alternating between 2 and 3 — when the supply runs low.)

6.5 Building and cascading rings: the PA3FWM ÷10, ÷5, ÷10, ÷6 chain

The practical craft of neon division is (a) building single rings of the ratio you need and (b) cascading them so each ring’s carry clocks the next. Any integer N is available: a ring of N stages divides by N. So ÷2 is a two-stage ring, ÷5 a five-stage ring, ÷6 a six-stage ring, ÷10 the classic decade — and a dekatron (Vol 5) is simply a ten-stage ring integrated into one envelope, dividing by 10 with a carry tap. To divide by a composite number you factor it and cascade: ÷3000 = ÷10 × ÷5 × ÷10 × ÷6. Because every stage divides exactly, the cascade divides exactly by the product — the errors do not accumulate, because there are no ratio errors to accumulate. What can accumulate is unreliability: a chain is only as reliable as its least-reliable stage, and a dropped carry anywhere corrupts everything downstream.

Building each ratio is just choosing the stage count and then applying the same Vol 4 design math to every stage regardless of ratio — the resistor and capacitor values follow from the tube’s Vs/Vm and the chosen current, not from N. A few practical notes on the common ratios:

Table 1 — ratios

Ratio	Stages	Build notes	Where it shows up
÷2	2	The minimum ring; two stages ping-pong. The octave-down cell — the flip-flop’s direct analogue. (The Philicorda’s ZA1001 octave stages are synchronised oscillators, not rings; §6.6.)	Counting octave dividers
÷5	5	An odd-length ring; carry taps the fifth stage. Pairs with a ÷2 to make ÷10.	PA3FWM 5 Hz→1 Hz stage
÷6	6	The sexagesimal workhorse — seconds→minutes and minutes→hours in any clock.	PA3FWM, and every 60-count
÷10	10	The decade; ten stages, or one dekatron (Vol 5) doing the same job in one envelope with a numbered dial.	Decade scalers, PA3FWM

Two rules make cascades behave. First, buffer the carry: a ring’s carry pulse is weaker than its input, so between stages insert the biased buffer lamp of §6.2/§6.5 (or, in period practice, a trigger tube or a small pulse transformer) so the next ring sees a clean, full- amplitude strike. Second, factor for reliability, not just arithmetic: ÷3000 could in principle be one 3000-stage ring, but nobody builds that — you cascade short, individually reliable rings (÷10 × ÷5 × ÷10 × ÷6) so each stage has few tubes to keep in margin and a fault is localised. Because each ring divides exactly, any factoring of N gives the same exact overall ratio; the only thing the factoring changes is how reliable and how serviceable the chain is.

Pieter-Tjerk de Boer’s (PA3FWM) silicon-free neon clock is the definitive worked cascade, and it is the one to study. Its timebase takes the 50 Hz mains as a frequency reference and divides it down to one pulse per minute through four cascaded neon ring counters — first ÷10 to 5 Hz, then ÷5 to 1 Hz (one pulse per second), then ÷10 to 0.1 Hz, then ÷6 to one pulse per minute. Four more ring counters (÷10, ÷6, ÷10, ÷3) then count the minutes, tens-of-minutes, hours and tens-of-hours for the display, so the whole clock is eight neon rings and not a transistor in sight.

The number that must agree across this whole series: 10 × 5 × 10 × 6 = 3000, and 50 Hz ÷ 3000 = 1/60 Hz, exactly one pulse per minute. And because the mains frequency in Europe is held to a tight long-term average by the grid operators, that 1/minute pulse inherits excellent long-term accuracy — the neon rings pass the grid’s timekeeping through untouched, dividing by an exact 3000. The clock keeps good time not despite the neon dividers but because they add no error to a good reference.

Two engineering details from that build deserve emphasis because they are exactly the reliability-vs-ratio lessons of §6.2. First, cascading modern lamps directly did not work: the carry pulse a ring of ordinary neon bulbs produces is too weak to reliably strike the first stage of the next ring. De Boer’s fix was an extra neon “buffer” lamp per counter, biased to sit just below its striking voltage so that a small carry pulse tips it over and it delivers a clean, strong pulse to the next stage — a neon amplifier, in effect. Second, the supply had to be stiff: he ran +151 V and +157 V rails, and later replaced a neon-reference regulator with a proper 150B2 cold-cathode stabiliser tube (Vol 8) precisely so that supply movement would not eat into the tube margins and provoke the drop/double faults of §6.2. And the honest coda — told in full in Vol 7 — is that even with binning, burn-in, buffers and a regulated rail, the clock became unusable after one to two years as the lamps kept ageing: “more parameters than just those two voltages play a role.” That is not a ratio failure (the arithmetic ÷3000 was always exact); it is a reliability failure, the count itself finally becoming untrustworthy. It is the whole volume in one anecdote.

6.6 The Philicorda: an organ voiced by synchronised neon oscillators

The most charming application of neon frequency division is not a clock at all but a musical instrument: the early all-valve Philips Philicorda electronic organ. The first production model, the AG7500, was launched at the beginning of 1963; the valve version of the 22GM751 that followed used the same scheme. Both are 49-key, single-manual organs whose tone generators were built, astonishingly for the era, from valve oscillators and neon tube frequency dividers, right at the moment the transistor was taking over electronic music. The full treatment of the instrument is in the sibling Philicorda dive; here the interest is in what kind of divider it uses, because it is not the kind this volume has mostly been about.

The architecture is the classic master-oscillator / octave-divider scheme. Twelve Hartley valve oscillators, each on half an ECC83 and tuned by a slug in its coil, generate the top octave, c♯⁴ to c⁵ (about 2217–4186 Hz). Each feeds a chain of neon stages: a first stage that runs at the oscillator’s own frequency and turns its sine into a sawtooth, then five stages (six for c) that each run an octave lower. That is 73 neon stages in all — 70 × ZA1001 and 3 × Z70U trigger tubes in the AG7500 (radiomuseum, SMEM), 68 × ZA1001 and 5 × Z70U in the 22GM751, whose 1967 service manual also lists one BA100 diode for each of the 61 dividing stages. A key’s contacts pick off the running outputs onto 8′, 4′ and 2′ busbars, and the timbre is shaped by filters after division.

The stages are not rings. The 22GM751 service manual describes each one as a Sägezahnoszillator — a sawtooth oscillator — and the circuit bears that out: a single ZA1001 charging a capacitor pair through 2.2 MΩ from a regulated 205 V rail, with a 30 pF trimmer that couples the sharp drop of the stage above into its cathode. Each stage is a free-running relaxation oscillator (Vol 2) whose own period is set a little longer than two periods of its input; the first kick after each firing is too early to fire it, the second fires it slightly early, and the stage locks at exactly half the input rate. That is synchronised division — the injection-locked kind of §6.4 — not counting.

The accuracy point survives in part. While a stage is locked its frequency is exactly half its input, so the dividers add no pitch error: every octave of a note tracks its master oscillator, drift and vibrato included, and the tuning of the whole instrument is set by the twelve coil cores. But the ratio is held by the lock, not by structure. Anything that moves a stage’s free-running period — a sagging divider rail, a 2.2 MΩ resistor drifting high, a leaky capacitor, a tube striking later as it ages — moves it towards the edge of its lock range. Past the edge the stage locks at ÷3 (a fifth below the intended note), alternates between ratios, or stops locking and free-runs, which owners hear as a rough “motorboating” note with its neon visibly flickering in time. A Dutch repairer reports “a 2.5-, 2- or 3-divider” at low supply. So the failure signature is not only the clean octave jump of a counting divider: a tired Philicorda stage can produce a pitch that has nothing to do with the note. The service manual’s tuning procedure accordingly ends with an instruction to re-set any dividers that have shifted after the oscillators are retuned. The instrument, its circuit values, the lock behaviour and the restoration of that neon bank are all set out in the sibling Philicorda dive.

6.7 Neon division vs. the other era dividers

It helps to place neon division among its contemporaries, because the same “counts, so the ratio is exact” logic applies to some of them and not others. Three families competed to divide a frequency before cheap logic ICs arrived, and they trade off very differently.

Table 2 — 6.7 Neon division vs. the other era dividers

Divider	How it divides	Ratio nature	Failure mode	Speed	Notes
Neon ring / dekatron	Counts N input pulses, emits one	Exact integer N (counting)	Drop/double a count → ±1 or octave glitch; jitter on edges	~1 kHz/stage	Self-displaying; HV; ages; wide-window switching tubes best
Multivibrator (free-running, synced)	Free-runs near f_in/N, injection-locked to the drive	Integer only while locked — can jump ratio	Unlocks → settles on a wrong sub-harmonic	~10 kHz–MHz	Cheap (a valve/transistor pair); the risky §6.4 kind
Blocking oscillator	Pulse-regenerative; one output per triggered cycle	Integer if reliably triggered (counting-like)	Missed/extra trigger → ±1	~kHz–100s kHz	Transformer-coupled; used in TV/timebase division
Bistable / flip-flop (later)	Toggles on every input; two toggles = one output	Exact ÷2 per stage (counting)	Essentially none until worn/faulted	MHz+	Valve then transistor then IC; ÷2^n by cascade; the eventual winner

The pattern is clear: the counting dividers (neon ring, dekatron, blocking oscillator that reliably triggers, and above all the flip-flop) share the neon ring’s headline virtue — an exact integer ratio — differing mainly in speed and reliability. The injection-locked multivibrator is the outlier and the cautionary case of §6.4: cheap and fast, but able to land quietly on the wrong ratio. The transistor flip-flop, and then the CMOS counter, eventually won for the obvious reasons — MHz speeds, low voltage, no ageing gas, no binning — but it won by being a better counting divider, not by inventing a new kind of accuracy. A ÷2 flip-flop and a ÷2 neon octave tube are conceptually the same machine: both divide by exactly two by counting to two. The neon version just does it in glowing gas at 150 volts, wears out, and is far more beautiful to watch. Everything the flip-flop taught us about trustworthy division — that counting gives you an exact integer, that the real enemies are reliability and jitter and not the ratio — the neon ring taught first.

6.8 6.x References

• P.-T. de Boer (PA3FWM), “A clock using neon lamps as logic elements” — https://www.pa3fwm.nl/projects/neonclock/ (the ÷10/÷5/÷10/÷6 mains-to-minute divider chain, buffer-lamp cascading fix, +151/+157 V rails and 150B2 stabiliser, and the honest ageing/reliability account).
• “Philicorda,” Wikipedia — https://en.wikipedia.org/wiki/Philicorda (twelve Hartley oscillators with frequency dividers; neon-bulb octave dividers; consistent warm tone).
• “Philips Philicorda AG7500,” Radiomuseum — https://www.radiomuseum.org/r/philips_philicorda_ag7500.html (tube complement: ~70× ZA1001 neon dividers, ECC83/ECL82/EZ80 valves; tube oscillators + neon frequency dividers).
• “The Organ That Forgot To Use Transistors,” Hackaday, 2026-07-03 — https://hackaday.com/2026/07/03/the-organ-that-forgot-to-use-transistors/ (popular account of the Philicorda’s neon-tube divider architecture).
• R. Dekker (dos4ever), “A Neon Ring Counter” — https://www.dos4ever.com/ring/ring.html (ring transfer mechanism, C0 sizing and the double-step limit, tube margins and ageing behind the reliability faults of §6.2).
• L. Small, “Neon Ring Counters” (2016) — https://lucsmall.com/2016/10/08/neon-ring-counters/ (binning, 48-hour burn-in and Vs/Vm characterisation — the reliability disciplines that make the exact ratio deliverable).
• J. B. Dance, Electronic Counting Circuits (London: Iliffe / New York: American Elsevier, 1967) — in the site’s reference library (canonical cold-cathode counting/dividing reference; multivibrator, blocking-oscillator and cold-cathode division compared).
• Cross-references: Vol 2 (relaxation oscillator, deionisation and the ~1 kHz ceiling, photosensitivity), Vol 3 (ZA1001/ZA1002 switching tubes and the Vs−Vm window), Vol 4 (ring transfer mechanism, C0 sizing), Vol 5 (dekatron as an integrated ÷10 ring), Vol 7 (drift, ageing and the full PA3FWM failure story), Vols 8–9 (stiff regulated supplies), Vol 10 (low-DA coupling capacitors for low jitter), and the sibling project ../Philicorda/ (the organ in full).