Mixing the weft divides a step and leaves a stripe
Worth reading first: A random error hides and a periodic one shows · A course is one thread and a warp is many · A weft stripe is counted in pairs of picks.
A warp is two thousand packages side by side and a weft is one at a time. A course is one thread and a warp is many made that the reason a knit shows a difference between two packages as a band and a woven cloth’s warp hides it: the warp averages its packages across the width, and the eye sees their mean. The weft does not. Every pick from one pirn has that pirn’s shade, and when the pirn runs out and the next is put in, the cloth changes shade on that pick and keeps the new shade until the next change. A weft bar is the difference between two pirns, laid across the whole width.
The trade has an old remedy, and it is not to make the pirns more alike, which the spread was never free showed has a floor no spinner can beat. It is to weave from more than one source at a time: two shuttles thrown in turn, pick and pick, or three or four feeders in rotation on a shuttleless loom. A mill does this precisely when it cannot trust its pirns to match, and the usual account of why it works is that it “mixes” the weft — which is true and says nothing about how much.
The arithmetic has three parts, and two of them are exact. The third is a question for the eye, and it turns out to be the one that decides which mixing is worth having.
A change of pirn moves one source of k
With k sources in rotation, a change of pirn changes one of them. The other k − 1 go on laying the shades they had, so in any stretch of cloth longer than a rotation the mean shade moves by a k-th of the difference between the old pirn and the new. Every step is divided by k, exactly. Nothing about the yarn enters, and nothing about how different the pirns are; it is a statement about averaging k numbers and changing one.
What mixing does not do is make the steps rarer. Each source still runs a pirn of the same length, but now it lays only every k-th pick, so it changes pirn every k pirn-lengths of cloth; there are k sources, and so the cloth still meets one change every pirn’s length. The steps are just as frequent and k times smaller. That distinguishes mixing from every remedy that acts on the yarn. A better yarn makes the difference between pirns smaller; mixing leaves the difference alone and takes a fixed share of it out of every step.
The obvious question is where the rest goes. A pirn that is a full step darker than its neighbours has not become a k-th as dark by being thrown on every k-th pick. It is still as dark, on the picks it lays.
The rest becomes a stripe
While the odd pirn lasts, one pick in every k carries its difference and the others do not. That is a pulse train — a stripe of period k picks with a duty of one in k — and it is a periodic error laid into the cloth on purpose. A random error hides and a periodic one shows priced exactly that kind of arrangement: a reed dent a tenth of a millimetre wide makes a streak, because a periodic error puts all of its strength into a few frequencies where the eye can find it. Mixing takes a step, which is aperiodic, and turns most of it into a grating.
The amplitudes of a pulse train are textbook, and they say something specific about each arrangement. Pick and pick, two sources one pick each, puts almost two thirds of the difference into a single harmonic, the fundamental, at a period of two picks — the finest periodic structure a cloth can carry, since nothing in a weft can alternate faster than pick by pick. More sources weaken the fundamental and spread what is left over more harmonics, and every one of their harmonics is at a coarser period than pick and pick’s fundamental. So pick and pick makes the strongest stripe and the finest one. Whether that is good depends entirely on whether the eye can resolve it.
Where the eye can follow a stripe
A stripe is a grating, and the eye’s response to a grating is something vision science has measured with care: the contrast sensitivity, the reciprocal of the least contrast at which a sine grating of a given frequency is seen. A figure’s streak is plainer from across the room used Barten’s standard formula for it, and this essay uses the same one, at a mean luminance of 50 candela a square metre — a lit room.
The frequency the eye meets is the stripe’s period divided by the viewing distance, in degrees. At 22 picks a centimetre, pick and pick’s stripe is 0.91 millimetres a cycle. From a quarter of a metre that is 4.8 cycles a degree, near the eye’s peak, where a contrast of a fifth of a per cent is visible. From half a metre it is 9.6 cycles a degree and the sensitivity has halved; from one metre, 19.2, at a seventh of the peak; from two metres, 38.4, where the sensitivity is 7 and a stripe has to modulate the cloth’s brightness by a seventh before anybody sees it.
So stepping back slides the stripe down the steep side of the eye’s curve. The same stripe is a fine visible texture at reading distance and gone at conversation distance, and the transition is fast: its threshold rises about thirteenfold between one and two metres.
How large a difference each stripe hides
Multiply each harmonic’s amplitude by the sensitivity at its frequency, take the largest, and the reciprocal is the least pirn-to-pirn difference at which the stripe is seen.
Pick and pick’s stripe hides a difference of 1.6 per cent from a metre and of more than twenty from two. Three sources in turn hide 0.89 and 4.2; four, 0.74 and 2.3. Every curve rises steeply once the stripe’s fundamental leaves the eye’s range, and the finer the stripe the sooner that happens.
Two and two is the case worth stopping on. A loom with shuttle boxes at one side only cannot throw pick and pick, because a shuttle has to come back to its own box and so every colour, or every source, is thrown an even number of picks: two and two is the finest rotation such a loom can make. Its stripe is the same duty as pick and pick’s at twice the period, so it reaches the eye at the same frequency from twice the distance. Two and two from two metres is pick and pick from one, and the 22.7 per cent pick and pick hides at two metres becomes 1.64 per cent. Boxes at both sides cost a shuttle for each source; what they buy, for mixing, is a stripe that disappears at half the distance.
A finer cloth hides its stripe nearer
The stripe’s period is k picks, so it is set in millimetres by the pick density, and the eye meets it as millimetres over metres. Pick density and viewing distance therefore trade one for one. A cloth at 44 picks a centimetre, mixed pick and pick, presents from half a metre exactly the stripe a 22-pick cloth presents from one, and hides exactly the same difference there. A coarse cloth at 11 picks a centimetre carries a stripe 1.8 millimetres a cycle, two and two’s at the finer density, and is seen from twice as far as the 22-pick cloth.
That makes pick and pick a better remedy on fine cloths than on coarse ones, for a reason that has nothing to do with the yarn. A poplin at forty or fifty picks a centimetre puts its mixing stripe at 0.4 to 0.5 millimetres, invisible beyond about half a metre for the differences in question, while a heavy twill at fifteen picks puts it at 1.3 millimetres and shows it to anyone standing near. The density was chosen for other reasons, and it decides how far away a mixing stripe disappears.
The same scaling holds for the step, but there it does nothing. A bar’s threshold on the generous reading is the eye’s best, and the eye’s best is reached at some distance for any bar wide enough to see at all; the pick density changes where that distance is and not the floor. The step’s floor is the same on every cloth; only the stripe’s ceiling moves with the density.
Random pirns rather than one odd one
The figures above follow one pirn that differs from the rest, which is the clean case. A real weft is a sequence of pirns whose shades are all scattered about a mean, and a cloth is a population, not a thread in this direction too. If consecutive pirns are independent with a spread , the difference between two of them has a spread of , so a single shuttle’s steps are typically and k sources’ are . Nothing in that changes the division by k, which is about averaging and not about where the shades come from.
What changes is the stripe. With every source carrying a pirn of its own random shade, the stripe in any stretch of cloth is the pattern of k shades repeating, and its amplitude is the spread among the k pirns currently running rather than the one difference followed above. For two sources it is still a pulse train of duty one half, with an amplitude that is the difference between the two current pirns — the same spread as a single shuttle’s steps. So pick and pick takes every step down to half and leaves a stripe whose amplitude is, on average, the full size of the step it removed, which is the hero figure’s picture: steps of half the height, and a band of alternation as wide as the old steps were tall.
The step has a floor too
The step is the other half, and it is harder to read through the eye, because a bar is not a grating. A bar seen across a piece is an edge, and an edge has some strength at every frequency, so it can be found wherever the eye is best. The reading taken here is the generous one: a bar is treated as one cycle of a square wave at whatever width puts its fundamental at the eye’s peak, so its threshold is a floor that no real bar can go below. At the peak’s sensitivity of 672, a step of 0.234 per cent is at that floor.
So a single shuttle shows a bar wherever two pirns differ by more than about a quarter of a per cent, and mixing over k sources raises that floor to k times as much, because every step is a k-th as large. That is the half of mixing that does not depend on distance at all.
Put the two together and each arrangement hides the lesser of what its step and its stripe allow. Pick and pick doubles the hidden difference at every distance, from 0.23 per cent to 0.47, and at none of the three distances is its stripe the limit — even at half a metre the stripe needs 0.52 per cent, just above the step’s 0.47. Beyond two sources the stripe starts to bind. From half a metre, three and four sources hide 0.40 per cent, less than two do; from one metre, three sources reach 0.70 and more sources add almost nothing, 0.74 at most; only from two metres does the gain run on to 1.4 per cent at six sources, where the step is still the limit.
The one-sided box loom loses most of the benefit up close. From half a metre, two and two hides 0.28 per cent against pick and pick’s 0.47 — a gain of a fifth over a single shuttle where pick and pick gains a factor of two — because its stripe, twice as coarse, is the limit before the step is. From a metre the two arrangements agree again. Which of these distances matters is a question about the cloth: a shirting is looked at from a hand’s breadth by the person wearing it and from a metre by everyone else; a curtain from across a room.
What mixing is good for
The arithmetic says what the trade’s remedy actually does, and it is narrower than “mixing” suggests. It cannot make the changes of shade any rarer, and it does not make the pirns more alike. It takes a fixed share of every step and turns it into a fine periodic stripe, and the remedy works exactly as far as that stripe is too fine for the eye at the distance the cloth is seen from. For pick and pick at an ordinary pick density that is almost always the case beyond a metre and often not at reading distance.
It also says why mixing is a remedy for small differences and never for large ones. The differences mixing hides are a few tenths of a per cent, from any rotation at any of the three distances. A difference of a few per cent — the kind of gap the habit of keeping dye lots apart suggests lies between lots — is not hidden by two sources or by eight: its step stays far above the floor whatever k is, and its stripe is visible within a metre. A mill mixes pirns within a lot and keeps lots apart, and the arithmetic agrees with the habit provided the within-lot differences really are of the smaller size; the measurement that would settle it is described at the end.
And it connects the bar to the other weft faults these essays have met. A slub finds the width of the cloth was a periodic fault in the yarn that the width turned into diagonals or stripes; the reed leaves its own mark was a periodic fault in the machine. Mixing is the one case where a periodic structure is put in deliberately, as the price of removing an aperiodic one, and the periodic-shows argument that condemned the other two is what decides whether the price is worth paying.
The model, and what it assumes
The cloth is a sequence of picks, each with the shade of the pirn that laid it, at 22 picks a centimetre. Mixing throws the weft from k sources in rotation, g picks each, and each source changes pirn independently when its own pirn is spent. The step a change makes in the cloth’s local mean is the difference over k; the stripe is a pulse train of duty 1/k and period kg picks, whose harmonics are taken in closed form and checked against the train’s variance.
The eye is Barten’s standard contrast sensitivity at 50 cd/m², with the field taken as the angle a cloth 1.4 metres wide subtends, up to sixty degrees. A stripe is seen when its strongest harmonic times the sensitivity at that harmonic’s frequency reaches one. The bar is read on the generous assumption that its edge is as visible as a square wave at the eye’s most sensitive frequency, which puts its threshold at π/2 over the peak sensitivity.
Required of it, not shown: that the stripe’s harmonics carry the pulse train’s variance; that the eye’s sensitivity peak between two and five cycles a degree, as a standard eye’s does; and that a rotation refuse fewer than two sources or a fractional number of picks.
What the reading leaves out
Summation is left out on the stripe’s side. The eye combines harmonics and repeats, and a stripe covering a whole piece is a little easier to see than its strongest harmonic alone says; the stripe thresholds here are slightly generous to the cloth. The bar’s side is generous to the eye, as stated. So where the step limits, the true hidden difference is at least what is given, and where the stripe limits it is a little less.
The shade of a pick is treated as a number. A pirn differs from its neighbour in count, twist, hairiness and dye uptake together, and each changes the cloth’s look differently under different light; the reading here takes the net difference in brightness as given. Nothing here says what a mill’s pirns actually differ by, which is the input every number above is scaled by.
The weave is not in it. A pick that floats on the face shows its shade more than one that is mostly buried, so a weft-faced cloth shows a weft difference more than a warp-faced one, and the stripe’s visibility should be scaled by the share of the face the weft occupies. That is balance entering a calculation it has so far stayed out of, and it is not done here.
Who knew which part
Weft mixing — two shuttles pick and pick, multiple feeders on shuttleless looms — is standard mill practice against weft bars, and keeping dye lots apart is older still. The contrast sensitivity function is Barten’s simplified formula of 2004. The Fourier series of a pulse train is classical.
What is added here is the accounting: every step divided by exactly the number of sources, with the changes no rarer; the remainder as a stripe of known harmonics at a period of k picks; the stripe read through the standard eye at a stated density, giving the difference each arrangement hides at each distance; and the finding that a loom forced to mix two and two hides little more than a single shuttle from half a metre, because its stripe is twice as coarse.
Still open: what a mill’s pirns differ by
Every number above is a threshold, and a threshold is only useful against the difference it is compared with. What is missing is the distribution of shade differences between consecutive pirns of one lot, measured on cloth rather than on yarn — the quantity the whole of the step-and-stripe arithmetic is scaled by.
It has a direct measurement. Weave a length from a single shuttle and read the brightness of the cloth pick by pick with a scanner at a fixed light; every step in the trace is one pirn change, and its height is one sample of the difference. A few dozen changes give a spread. The prediction is that within a lot the spread sits at a few tenths of a per cent — the scale where a single shuttle’s bars are just visible and pick and pick’s are not — and that between lots it is several times larger, where no rotation helps. If within-lot differences turn out larger than about half a per cent, pick and pick is not hiding the bars it is credited with hiding, and the habit is working for some other reason.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A finish spends a spread before it spends a mean — both name appearance, coefficient of variation, population
- A moiré is a vernier, and it magnifies the error too — both name coefficient of variation, period, population
- Two sheers make a moiré that walks with the viewer — both name coefficient of variation, harmonic, period
- A bundle is weaker than its threads — both name coefficient of variation, population
- A chenille is a yarn that is already a fabric — both name coefficient of variation, population
- A cloth cannot be more even than its yarn — both name coefficient of variation, population
Named objects
A flat tag is an object no other essay names yet.
AppearanceCoefficient of variationHarmonicPeriodPopulationStripe