A computer works by counting the number of times the magnet passes its sensor, and calculates speed/distance using the circumference number you enter for a particular tire size. You could put the magnet 6in from the hub, or 12in from the hub, but the magnet is only going to pass the sensor ONCE per revolution. All the rest will depend on how accurate your circumference measurement is.

FWIW a cadence computer uses the same technology, but reads RPM directly, without needing to calculate anything further.

This is correct but the the magnet farther away from the hub is moving faster past the sensor. It appears (according to Sheldon Brown) that the counts are more reliable if the magnet is closer to the hub.

The issue with the distance from the hub is that the faster speed farther from the hub can cause some of the rotations to be missed, which means the counts are lower than they should be.

Keep in mind that the gap between the magnet and the sensor can't be too large. (The closer, the better.)

i got the magnet and sensor as close as i can get them together within reason

Another article on Sheldon's site was news to me but makes sense since all magnets are dipoles (have two poles), magnet orientation can make a difference and cause speed dependent errors. See Cyclometer Errors Caused by Magnet Orientation.

Most of the computers I have used say to set the pickup (sensor) at 5mm from the wheel magnet. I set mine to 3mm, and have never had a problem with 'missed counts. That is not to say that this will work for everyone. I also put the magnet out near (3-4in) the rim, to act as a weight for wheel balance. It has worked well for me.

I do the same ride a few times a week. It's around 27 miles. I am always fascinated when my Cateye Strada Wireless reports the distance as 27.11. The largest deviation I can get is a couple of 27.13 distances. Most are spot-on at 27.11 or 27.12. That's EXTREMELY consistent and makes me really like their product.

0.01 miles is 52.8 feet. That's a small variation compared to 27 miles. It's surprising that your ride path doesn't vary more than that. That is, it's a bit weird that they are always so close. And fact that you get some variation doesn't mean that the variation is due to inaccuracies in the computer.

All bike computers (ignoring the GPS ones) just count wheel rotations. There is no reason that the Cateye would really be any better at doing this simple task than any other computer (of reasonable and comparable quality).

??

In terms of balance, it really doesn't matter where the magnet is placed. Anyway, you'd reduce balance "issues" with the magnet closer to the hub.

njkayaker: If you spin your wheel slowly, until it comes to rest, the light side will be up. If you put weight on the light side, it will be closer to being in balance (hard to get it perfect). Placing the magnet (weight) closer to the rim puts it where it has more effect on the imbalance. I have done this on several wheelsets, and have had good success this way. I even tripled my presta valve lock nuts in conjunction with magnet, and got that wheel nearly perfectly balanced. It works.

You mentioned quality ... but even the cheap ones merely count wheel rotations too, which is a simple task that even the cheapest cyclecomputer can do.

However, there are a few quality related things that can make a cyclocomputer less accurate --

1. the magnet/sensor setup may not catch every revolution. This is especially likely if they're not installed properly -- if the magnet is too far from the sensor, stuff like that. But the better quality ones often have better magnets and more sensitive sensors and so are more tolerant of bad installations.

In some cases, some percentage of revolutions may be missed when going extra fast but things will be fine when going slower.

Corrosion in the contacts between the unit and it's base could lead to a situation where some revolutions are missed as well. The better cyclcomputers could try harder to keep water out of there or use a metal that doesn't corrode.

2. the magnet/sensor setup may catch extra revolutions. This is unlikely for wired setups (though not impossible -- I can see a bump being detected as a rotation in some setups and cases) but quite likely for wireless setups.

A cheap sensor may "vibrate" occasionally and pick up one pass of the magnet as two revolution.

Some wireless computers really freak out when they pick up noise, saying you're going 99 mph or something along those lines. (And it could go the other way as well, interference making your bike seem slower, especially on the better ones that try to code their signals so they only pick up their sensor.)

Really, if you want it as accurate as possible, you probably don't want it to be wireless (unless it's a GPS based unit, of course.) Some may work perfectly in most situations, but eventually they too could run into something they couldn't handle.

I said "quality" not "expensive". While there should be no reason to need to spend a lot on a computer, there probably are some that are so cheap to be garbage.

About 100% of the counting is being done at the "magnet/sensor".

It's this that I was getting at with "quality".

I was going to list some of the other problems but I hoped that the link to Sheldon Brown discussed them.

Wireless appears to be generally just fine (except for "max speed", once in a while).

There isn't any reason the coded wireless (eg, ANT+ units) have any problems (even with "max speed").

Note that, sometimes, the devices you carry (eg, LED flashlights) can cause interference. This can often be fixed by moving the device away from the wheel transmitter and head unit.

I said "cheap", not "expensive".

Cheap means two different things -- low in cost and low in quality. So "cheap" was an appropriate word to use there.

And there are some that work really well, but corners were cut on the magnet/sensor part so it's very sensitive to proper installation. The Schwinn 17 function one that's $10 at Amazon comes to mind -- good cyclocomputer, good functionality, great price, but the magnet and sensor need to be very close to perfect for it to work.
OK, that's wrong, unless you're talking about wireless units, then maybe.

The magnet/sensor either creates a small voltage (due to the Hall effect or magnetic induction) or opens and closes a circuit, depending on design. There is no counting here, it's just a change in voltage or resistance. The computer part does all the actual counting -- it counts these voltage or resistance changes, then does the appropriate math to give you a nice mph or kph display, to keep track of distance, etc. Sure, it's possible to do the calculations in the sensor unit, but why for a wired unit?

There's several things where this can go wrong some of the time -- the voltage/resistance change can be so small that the computer can't detect it, the magnet may not be close enough to the switch to even reliably trigger it, the "pulse" may be so quick that the computer can't detect it, etc. Better quality units generally have larger margins of error for anything that's not perfect.

If the unit says you're going 99 mph for an instant, then it's likely adding up miles as if you were going 99 mph for that instant. So if the max speed is way off, the odometer reading will be inaccurate -- but exactly how inaccurate depends on just how often that happens and how long it lasts, and what the computer does with these values (it could say "99 mph? No way! Just keep adding the last (valid) value seen" -- not that I've ever seen this done.)

It depends on the specifics, which I can't claim to know about all units.

Some (most? all?) wireless cyclocomputers just send a pulse of some sort each time the tire rotates. If this pulse isn't received, then the cyclocomputer thinks you're going slower than you really are. If multiple pulses are received, it thinks you're going too fast. Interference (possibly even from the sensor of the guy next to you) often creates extra pulses to be detected with most low cost/quality units which will do nothing to verify that this is a valid pulse.

It's also possible for the wireless sensor to not just send a pulse, but send the time since the last pulse encoded in the pulse. If they did this, the computer could ignore a few missing pulses and assume you're still going the same speed and the accuracy would still be pretty good as long as not too many pulses were lost. I do not know how many cyclocomputers do this, but I would expect the better quality ones to work this way, and the low quality generally seem to not work this way.

The good quality wireless units send a serial number or something along those lines (this is likely what you were getting at with ANT+ units) with each pulse, so the computer knows that this is a valid data point from your sensor. However, they're still not immune to interference -- if the pulse is garbled by remote interference (possibly even another sensor owned by the guy next to you that transmitted at exactly the same time), it won't validate and then the computer will have to do whatever it does about lost pulses.

Absolutely.

If this is the case, since the magnet/sensor is really where the important stuff (accurately registering ticks) for counting goes on, then this cheap computer would not be a quality one.

What I'm saying is that the accumulation (the counting) of ticks is easy and it's the generation of each wheel rotation that is hard. It's registering the ticks and transmitting them to the head unit that is the critical part.

All of the counting and distance computation occurs in the head unit for all computers, whether or not they are wired or wireless.

Yes, and personally, I don't want to debug problems with lesser quality units!

As long as the transmitter is very reliable (which has to be true for all "quality" computers), then the only problem with coded units is lost ticks.

If you lose ticks one in a while, I doubt you'd notice it (especially if the computer uses a moving average). The coded computers won't register spurious (ie, extra) ticks. Keep in mind that "max speed" problems are evidenced by ridiculously high maximum speeds (these are caused by extra spurious ticks that coded transmitters avoid entirely).

I think all of them just send a wheel rotation tick. I'd guess the computers (the better ones) compute a moving average for the speed. That would smooth-out any infrequent lost or spurious ticks.

The interference occurs with in wireless transmission. It can't (generally) occur at the sensor because the distances required to trigger a pulse there are tiny (it's possible that a large field could trigger a pulse but I have no idea what kind of thing that you'd be riding next-to would be able to generate that kind of pulse).

Interesting idea but I don't think they do this. The head-unit computer can do sanity checks of the speed. The bicycle isn't likely to change its speed by half in one wheel rotation. I believe a 700c wheel turns 2 times per second at 10 mph.

Yes. Keep in mind that the coding eliminates extra spurious pulses.

The only way to avoid problems with lost pulses is if the lost pulses are rare. If they are rare, it should be easy to smooth out the speed. If they are not rare, then the problem should be obvious!

I'm looking for a little reality check on my roll out measurement. I'm using a Cateye Strada wireless computer. Today I went on a 9 mile ride and compared the Strada's distance to a Cyclemeter app on my iPhone. The Cateye came out about .25 mile longer than the GPS app.

My bike has Bontrager 700 x 35c tires. According to the Cateye circumference table this works out to 2168mm which is how I set the computer. However, I did a roll-out measurement after my ride and it came out to about 2180mm which is closer to 700 x 38. Is it possible that I have a tire marked at 700 x 35c that has a 218cm circumference?

The difference is only 1/2 of one percent, or a radius difference of 2 milimeters.

The margin of error in your rollout test is likely larger than the discrepancy you see.

Comparing to the GPS, that's still only 2% ... pretty good.