DCC occupancy detections false positives

Well, I would more or less agree with this analysis _but_ the detector is inductively coupled, which means it will be much more sensitive to higher than to lower frequencies.

My own block occupancy detectors work off the voltage drop from two diodes in series, which is used to feed an optocoupler[1]. I find it's very reliable: no spurious detections even though it will detect a firmly-pressed-down dry finger (across N-scale track) or an inactive function decoder drawing a fraction of a milliamp.

I've got lots of big runs of wire in parallel and quite a bit of EMI (I had to shield a different part of my circuit with paper-wrapped tinfoil) and I see no problems.

[1] circuit, best I can do in ASCII art:

from booster output | | 5V +--------+-----------+ ------+--------- | | | | | | | +-+ | | +-+ | | ----- +---+ | | | | / \ \ / | | +-+ +---+ ----- +-+ | detector output | | | +--+-------------- | | | | / to microcontroller | | | __ |/ ----- +---+ +---+ ,. /| |\ / \ \ / \ / ' ~~ | V +---+ ----- ----- | | | | | | | | | | | | | GND +--------+-----------+ ---+-------------- | | to track

This has a minimum number of parts (and solder joints) per block: one bridge rectifier, 1/8th of an octal optocoupler, 1 ordinary resistor,

1/nth of a SIL resistor array for the pullup, and of course one microcontroller pin. I can fit 21 of these on one modest PCB along with the microcontroller, circuitry to control a dozen point motors, etc.

I built the booster myself and its output voltage is higher than the

12V nominal track voltage to compensate for the 0.7ish per diode forward drop.

Do you use current-based reversing loops? I had a couple problems with false triggering early on, when the occupancy sensor was on the track feeder hooked to the reverser. If a train entered the block from the opposite end, the occupancy detector would trigger for a bit (long enough to cause the computer to throw that switch thinking a train was coming), even though the sensed block was seperate - throwing the relay in the AR-1 triggered the detector.

The key thing with these detectors is that you MUST be certain that there is nothing but train that can pull current from the sensed block. Breaking the block up so that the sense block and the power block didn't start in the same place seems to have solved it. *

Just to amuse you, I once had a calculation of shorting-current in German (real) catenary performed. You start off with 15 000 Volt at

16 2/3 Hz (stupid frequency, but anyway): short it at the end which is far away from the supply point and you get a short blast of some 10 000 Amps - just from two "wires" running not even directly next to each other (a few meter distance) for some few kilometers. You have a non-neglectible capacity in there and a non-neglectible inductance.

Now, scale it down: the feeder wires run next to each other (a milimeter or two of distance), for 50 or 100 ft. and you have frequencies in the kiloHz range. You won't get the 10 000 Amps blast, but you will have some "nice" side effects. Not much, but probably enough, especially if you add some more stuff:

- the wires may be running parallel next to each other (and not be drilled around each other) which *might* give you some effect of radio-frequency interference,

- the wires might be soldered together for length and plugged into the booster, they will be soldered to the rails. As we're talking about AC signals, each change of AC resistance might produce some little reflection which lowers signal quality,

- as described in the original post, this happens when there are lots of people in the room which means higher air humidity. With some of the afore mentioned effects this might just be enough to "trigger" the false detection

- many people also means there might be a cell phone (or dect phone) disturbing the signalling, or there might be lots of wireless controllers, probably even somebody leaning against the benchwork causing a minor change in geometry which allows some leak current...

My suggestion would be to (1) check the cables and tracks, especially if this happens in certain places only, (2) try changing the ventilation of the room for the next OP session, (3) move the detectors as close to the track as possible in terms of cable length, (4) if you exchange cables somewhere, try to use shielded (computer data) cables and ground the shields. Surely, this is a very last resort as it is major work. There were other suggestions in this thread which are easier to follow ;-)

Good luck and lots of fun...

PS: This post is based on the (educated) opinion of the author.

On 5/20/2008 6:00 AM Bernhard Agthe spake thus:

Interesting: I'm a little confused by some of the details (i.e., what exactly is being shorted to what) but get the general picture. Of course, this spectacular "blast" only occurred on paper and pencil. I'm curious just how you derived this calculation.

Specifically, and here I'd like to relate this to the original discussion I was having, how do you calculate capacitance between conductors in air? And can you tell us what the capacitance of typical HO track would be, say per foot, in microfarads (or, more likely, in picofarads)?

I'm still skeptical that any such capacitance could have any measurable real-world effects at the relatively low frequencies found in any DCC system (and even more skeptical of your claim at a nearly subsonic frequency), but am open to being persuaded.

Hi,

They demonstrated a short between the overhead wire and the tracks at a point far from the feeder. Similar things would happen in long-distance power lines ;-) It all starts out with rather simple "school-physics", but gets wierd (sorry, mathematics is my weak point ;-) at the moment of the short because you have the following problem: there is a very sharp voltage change. Because the system has a non-zero capacity and inductivity, there is a compensating effect in the form of a current overswing. As it's a few years back, I would have to look up my university records which are back at my parents place...

(OT:) Actually, in case of AC, there is a point where the voltage reaches zero every so many microseconds, which is not the case in DC train systems. While the voltage is lower, usually ~3000 Volt, the big problem there is the engine's main switch. They do have severe arcs and sometimes the switches burn "on" - which will mean the unlucky engine may have its wheels soldered solidly on the rails (probably with the arc jumping about happily inside the engine). So lineside "short detection" is a real task in case of DC power ;-) (/OT)

Capacity is not a matter of enclosed conductors. It also works with air as the isolator medium. Basically you can use the formula from your school's physics classes (sorry if I kind of expect you to have the same school background as I had, but my physics classes were really good, just too short):

C = eA / d

with C=Capacity, e the dielectric factor (describing the isolator), A the facing area of the plates (wires) and d the distance between the wires. You can look up the general scheme at [1].

[1]

In fact I do agree we're talking about very small capacities (and inductances), but in a case where the signal is already of a bad quality, the distortion of such a small capacity may be enough.

As I recall my Comm-Tech courses, there are a few more problems, namely

- different frequencies having different "running speeds" - which is a problem in case of square waves, Gauss-Impulses were suggested as better, even sine-waves.

- compare the wavelength to cable-size: 10kHz makes about 30km wavelength, so we can sure discard this and assume DC calculation rules (as to AC rules where you have to care for phase-shift),

- the receiver filter may be matched to the impulse form, but it doesn't take into account the noise received and the transfer function of the wires and tracks,

- Usually the cables on model railways are not shielded while computer cables are usually shielded even if they're only a meter (3 ft.) long. With small signals, there may be enough noise received...

That was why I suggest (and concur with several other suggestions here) the OP should check the "simple" things first - sensitivity, room ventilation, the occasional "cold" soldering, a hidden leak producing some minor current flow, interference by cell phones... If those simple things don't help, the use of shielded cables for long-distance routing may or may not help - as the placement of the detectors may or may not have an influence...

Probably not, but I would not rule this phenomenon to be completely absurd. I have seen fuses go just by plugging a cable into a wall outlet. The cable was too long and had been coiled in a place - the resulting inductivity was enough - after I had re-cabled the whole place, everything was fine. So eventually, the residual capacities may grow large enough to actually matter - as I tried to demonstrate in the Train-short-power-blast example above. Go figure what happened when they encountered something like this for the first time 100 years ago ;-)

Have some fun...

I want to thank everybody who wrote suggestions and discussion. I'll try to answer all the questions and share my proposed line of attack for your comments. Whatever I do, I really would like to find some test that I can apply by myself. Having to wait for the next operating session at my house doesn't appeal. Nor does trying to debug the detectors while the guys are trying to run trains.

My layout is in a 20' x 24' room. I have a dehumidifier, but haven't turned it on yet for the season. In the winter I use a humidifier, but in spring and fall I don't need either.

The block being detected is 14' of straight track followed by a rising curve about 15' long that passes over the straight track. Both ends are gaped. Power from the booster is distributed using 12 gauge wire power bus running roughly parallel to the track. When I decided where the detection block was, I disconnected all the feeders on one rail from the bus and connected them to a sub-bus that is about 18' long. The detector circuit board is near one end of the sub-bus. The wire from the detector to the bus is about 1' long.

I measured resistance between the rails using an auto-ranging ohm meter; it measured as an open circuit. I realize that's a DC measurement, but I don't have any way of measuring at DCC frequencies.

It seems that some people refer to one pass of the track lead through the coil as one-half turn while others call that one turn; I'll use the latter. If there's a difference in configuration between one-half turn and one turn it escapes me.

I am planning to have 5 detection blocks. So far I have two configured. Only one of the detectors is showing the false detection. That presents several ideas. (1) I could switch the detectors and see what happens. (2) I could replace the detector that is giving the false positive with one that I haven't used yet. (3) I could try putting both detectors on the same block in series.

I adjusted the sensitivity to detect the caboose by experimenting with number of turns through the inductor coil and the resistance across the track. With one turn it could detect 2K ohm. With two turns it detected 5K. I decided that was sensitive enough and wired 9.6K resistors between the wheels of two axils on the cabooses. In general, I'm still using plastic wheels, so I purchased new metal wheel sets. I confirmed detection of cabooses.

I tried the dry finger test. The false positive detector switched states and the other did not, confirming that the first detector is more sensitive.

That's all I can think of now. I'd appreciate comments on my ideas of things to try or other ideas.

I'd like to take the thread off on a new tangent. I notice that the commercial occupancy detectors are pretty expensive on a per block basis, around $5 or more. What is the cheapest detector circuit that could be built? I've been checking out the various current sensors available and I still haven't found anything that could get the cost down to $1 per block or less.

Sounds to me like you now have some sound troubleshooting procedures which will probably enable to solve your problem.

Good job! Peteski

Well, not in dollars, but my circuit (see my previous posting) can be made as cheaply as this:

1/4 quad optocoupler eg TLP521-4 L1.08 27p 1x bridge rectifier (4 diodes) eg Fairchild DF08M 15.2p 2x resistors eg 2% 0.1W L0.001 0.2p ----- 42.4p

which is a little under $1. I'm assuming that you're having lots and lots of these - at least a hundred - so I used Farnell, a serious UK electronics supplier, and assumed (for example) the 50-off price for resistors and the 25-off price for quad optocouplers. You need to budget for some board to mount it on too which will at a guess will add 25-50p to the cost per block. Also these prices don't include VAT (UK sales tax).

Resistor values I use are 56R on the optocoupler input, but this will depend on the desired sensitivity and the voltage drop provided by your actual bridge rectifier. I use a 100K pull-up on the output.

In my setup I then feed the outputs to PIC18F458 microcontrollers for scanning and multiplexing; the PIC has quite good properties on its inputs which means I can feed the output from my circuit straight into its general IO pins as they have an excellent high input impedance of

1MR or so, don't seem to mind wavering levels, and which I can program to debounce the incoming signal before reporting detections.

In my setup, if you count the cost of the microcontroller too, and the custom PCB to mount it on, and all of the gubbins for reprogramming the microcontrollers, talking to them from the computer, etc. then it works out as a lot more than $1 per block :-).

But I don't know what you want to do with your output. If you need an output which remains constant rather than flickering between `present' and `not present' at 10-20kHz (due to the NMRA DCC squarewave, with the detection circuit only working during one half of each cycle) you need to at least add a capacitor (say, 10nF at 5.3p) and you'll probably need to use something to provide a hysteresis, eg a schmitt trigger 74HCT14 (10.5p for 10-off but you get six to a chip) - that will give you a TTL-level output for another 7.05p per block if you don't bother to include a decoupling capacitor. Total: 49.9p which is pretty close to US$1 :-).

Of course all of this more or less assumes you have a suitable 5V power supply and want an output you can feed to something resembling a TTL input. And as I say I haven't included the cost of some matrix board for mounting, which will be a significant fraction of the cost.

I think the real cost of anything along these lines depends on what you're going to do with the output from the detectors.

I'm afraid I was unable to make heads or tails of the attempt at an ASCII schematic. Could you perhaps draw up a version in Paint that could be posted somewhere?

...

Did you look at the schematic in a fixed width font ? I can't be bothered to draw another diagram but here's a screenshot of what it looked like for me:

The four ordinary diodes on the left are most easily found together in a bridge rectifier: short out the rectifier's `+' and `-' terminals and connect the two `~' terminals to booster output and track. (This ties the two inter-diode midpoints together, which is not a problem.)

The thing with a diode, a little arrow, and part of a transistor is the optocoupler. The rectangular boxes are the resistors.

I do have a more useful schematic at home that I could take a digital photo of and put online if it would help.

OK, the diagram at that link looks better. I can make it out now. Thanks.

Those are only rated at 1.5A. I doubt that I'd trust anything rated less than 5 amps, and preferably more just to be safe. My old brass locos could easily pull 2 amps when heavily loaded. Double head a couple and... well, you get the idea. Also I sort of wonder about the reliability of using standard rectifier diodes or diode bridges on the higher frequency DCC power. I suspect some of those bridges designed for use on 50 or 60 Hz power may not have a fast enough shutoff time when the signal reverses.

The OP didn't specify their scale IIRC. Mine's N so 1.5A is plenty.

The best thing about working with the proper electronics industry is that you can just look up this kind of question. Farnell, for example, provide a link to their copy of the datasheet PDF alongside each part listing in their web search system.

The 1.5A is average rectified forward current, so since you've got two pairs of diodes each of which only needs to carry the load half the time, you're up to 3A average. That'll be primarily thermally limited so if you want more you need to buy a more expensive part in a bigger package. The occasional surge it will allegedly cope with is 50A which I hope is fine even for a large scale setup.

My actual layout uses a higher-rated bridge rectifier, SB601 (Farnell

7278276) and similar, rated for 6A average. I chose such a chunky part because I wanted to keep the forward drop relatively independent of load so wanted big junctions with good thermal properties. The OP was trying to optimise for cost in which case those parts at over $2 each aren't the ideal answer.

But even so I've not had any trouble with shutoff times. I've conducted tests running 3A through a single block, and everything has still worked just fine (my tests included looking at the on-track waveform shape as seen on a scope, under heavy load). My test load - a nominally 2W rated 5R resistor - did get very hot as was to be expected from pumping 40W into it!

Hi,

this is slightly OT, but I'll try anyway.

Yesterday I finished my control pult. It connects to the layout via two

36-contact Centronics plugs. Most of the contacts are used already, so any occupancy detection I want to add has to be in the control pult only ;-) Since I use a DC system with two throttles and a common return, this is an interesting challenge ;-)

Anyway my idea is to use a high frequency (some KHz) which I apply to the common return via some capacity. Detect would be on the cables for block power supply, again coupled via capacitors. The detectors would be used to back-light the switches for each block, indicating a train in that block.

The interesting part is, that the signals would have to travel for up to

4 meters in quasi-parallel wires, through track and train and still be detectable without interference between neighbour lines... So to say, the wires themselves will provide some capacity - plus the coupling into and out of the track power system by capacitors leaves very low tolerances anyway...

The "sender" would basically be an oscillator circuit and the receiver would sense any HF voltage parts and use them to power a LED.

Does anyone have any experience with such a system - I'm a hobby electrician, but have little to no experience with HF stuff...

Actually I'm pretty sure, the principle will work in other than my special conditions, even with DCC, if the parameters are right ;-)

Thanks in advance for any input on the matter..

Ciao...

Actually on reflection I think this isn't true and the 1.5A figure is supposed to be for the whole part, rather than per diode.

So if you have locos which pull 1-2A probably you'd be better off with a bigger part, perhaps something like KBU4D at 57p which entirely blows the US$1 per block budget. Instead you can just use four ordinary diodes eg 1N5400G at 5.9p each, which is twice as much soldering but much cheaper.

The data sheet I read referred to the rating as "continuous current", or the maximum amount of current you could put through it on a constant basis without stressing it abnormally. Not to be confused with surge current which is a higher amount a device can stand for a very short, non-repetitive spike. At my job, which is repairing industrial controls, we normally order parts from Mouser, Allied, Newark (the U.S. counterpart of Farnell), and Digi-Key.

In the case of a bridge of any sort, the continuous current is the amount you have to expect any individual device in the bridge to be able to handle. So yes, the 1.5A rating would be per diode. I think a normal rule of thumb is to use a device capable of 150% of the maximum current you expect to be drawn, and 200% of the voltage. This correlates with the parts used in the equipment that I repair on a daily basis.

Checking Mouser's web site I see they have some discontinued STBR406

600V/5A bridges they're unloading for $.99 ea. Those would work, if I had the spare cash to spend ATM.

There's actually a US patent on this idea/design. the abstract is vague enough that it would cover your design, too, I think. I didn't look at the details.

For a survey of occupancy detection ideas, look at

HTH

Hi,

Thanks for your input on the topic, I do appreciate it.

Probably I will go for the approach of inserting a high-freq. voltage and trying to detect it, but I'm still not sure whether to use transformers or (probably) capacitors - I will report here if I find any interesting things ;-)

Ciao...