On Mon, 27 Jul 2009 08:51:56 -0600, "Not Really Me"
<scott@validatedQWERTYsoftware.XYZZY.com> wrote:

>Meindert Sprang wrote:
>> "Not Really Me" <scott@validatedQWERTYsoftware.XYZZY.com> wrote in
>> message news:7cral9F28uaiiU1@mid.individual.net...
>>> Don't try this with SPI, the distance is too great.  I think it is
>>> unlikely to work over I2C for the same reason, but I have less
>>> direct experience with that.
>>
>> As Vladimir already pointed out: IDE, GPIB and Centronics were also
>> single-ended at TTL level and all worked fine over at least 2 ft of
>> flatcable. The SPI cable from my programmer is also 1 ft and I have
>> never experienced any problems. Looking at the signals with my
>> oscilloscope does not show any ringing, overshoot or reflection that
>> matters.

Typically in flat cable busses every other connector carries some
signal and every other conductor is connected to a signal ground. With
a signal ground on both sides of the signal conductor, there is a well
defined characteristic impedance on the line (unless there are
metallic surfaces close to the cable), so it is possible to build a
matched system with no over/undershots. 

However, any external magnetic or electric field can induce noise to
the connection, since the signal and ground wires are not twisted
around each other that would cancel some of the induced noise.

The construction also requires  that no other DC or mains frequency
currents are flowing in the signal ground conductors. In some cases
this has required strapping together the frames of the CPU and the
disk cabinet with 5-10 cm wide copper braids :-) in order to avoid
problem caused by potential differences in the mains neutral wiring.

>What speed SPI?  We used ADCs ADuC8xx series and had horrible results at 1 
>ft until we used twisted pairs.  It's been a few years but I think it was 
>set for 10Mbps but was still flaky at 250Kb, the lowest speed it had.  Only 
>choices were 1M, 10M and 250K.

While the input and output pins are labeled as "TTL compatible" on
most modern digital chips, have you actually looked at the input and
output pin characteristics for input and output impedance and how well
it matches the characteristic impedance of the flat cable (typically
slightly below 100 ohms) ?

Paul


Meindert Sprang wrote:

> "Not Really Me" <scott@validatedQWERTYsoftware.XYZZY.com> wrote in message
> news:7d5t0gF2amsl4U1@mid.individual.net...
> 
>>Well, he said 4 ft, so that would be more trouble than 1ft.  I did have
>>either a 12 or 20 ft Centronics cable, but that is a slow interface.
>>
>>What speed SPI?
> 
> 
> 1.8 MHz.
> 
> 
>>We used ADCs ADuC8xx series and had horrible results at 1
>>ft until we used twisted pairs.  It's been a few years but I think it was
>>set for 10Mbps but was still flaky at 250Kb, the lowest speed it had.
> 
> 
> I think something else must have been wrong. I just ran a 5MHz square wave
> through 3ft of speaker cable, driven from a 50 Ohm generator. The signal
> comes out good. Overshoot is 17% and no degradation of risetime (better than
> 5ns). when terminated with 100 Ohm, the overshoot disappears. The same
> signal driven from a TTL gate shows 25% overshoot.
> 
> So with a bit of a decent driver and termination, the OP can get good
> results with simple single ended terminated wire and a 50 Ohm driver. RS-485
> or CAN would be overkill.

You are correct.
I have the Altera Byteblaster on the 10 ft cable. Doing the SPI both 
directions by bit banging at ~1.5 MHz.  No problem.

> 
>>I would still not recommend SPI or I2C for this.  It is always better to
>>used something intended for the use that you have and neither of these 
> does.
> 
> And when is something intended for the use? IMO anything that gives a good
> reliable result.

It depends. Once I was surprised to see the remote sensor with the 
Dallas 1-wire interface operating OK over ~15m of the twisted pair; 
however I've also seen the I2C not working on the same PCB with the SMPS 
because of the problematic layout.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On Jul 27, 4:51=A0pm, "Not Really Me"
<sc...@validatedQWERTYsoftware.XYZZY.com> wrote:
> Meindert Sprang wrote:
> > "Not Really Me" <sc...@validatedQWERTYsoftware.XYZZY.com> wrote in
> > messagenews:7cral9F28uaiiU1@mid.individual.net...
> >> Don't try this with SPI, the distance is too great. =A0I think it is
> >> unlikely to work over I2C for the same reason, but I have less
> >> direct experience with that.
>
> > As Vladimir already pointed out: IDE, GPIB and Centronics were also
> > single-ended at TTL level and all worked fine over at least 2 ft of
> > flatcable. The SPI cable from my programmer is also 1 ft and I have
> > never experienced any problems. Looking at the signals with my
> > oscilloscope does not show any ringing, overshoot or reflection that
> > matters.
>
> > Meindert
>
> Well, he said 4 ft, so that would be more trouble than 1ft. =A0I did have
> either a 12 or 20 ft Centronics cable, but that is a slow interface.
>
> What speed SPI? =A0We used ADCs ADuC8xx series and had horrible results a=
t 1
> ft until we used twisted pairs. =A0It's been a few years but I think it w=
as
> set for 10Mbps but was still flaky at 250Kb, the lowest speed it had. =A0=
Only
> choices were 1M, 10M and 250K.
>
> I would still not recommend SPI or I2C for this. =A0It is always better t=
o
> used something intended for the use that you have and neither of these do=
es.

The problem is the rate of rise and fall, not the actual clock
frequency. If the rise time on the 250kb clock (or data line) is
5nanosec then one is actually looking at 100Mhz timing.

"Not Really Me" <scott@validatedQWERTYsoftware.XYZZY.com> wrote in message
news:7d5t0gF2amsl4U1@mid.individual.net...
> Well, he said 4 ft, so that would be more trouble than 1ft.  I did have
> either a 12 or 20 ft Centronics cable, but that is a slow interface.
>
> What speed SPI?

1.8 MHz.

> We used ADCs ADuC8xx series and had horrible results at 1
> ft until we used twisted pairs.  It's been a few years but I think it was
> set for 10Mbps but was still flaky at 250Kb, the lowest speed it had.

I think something else must have been wrong. I just ran a 5MHz square wave
through 3ft of speaker cable, driven from a 50 Ohm generator. The signal
comes out good. Overshoot is 17% and no degradation of risetime (better than
5ns). when terminated with 100 Ohm, the overshoot disappears. The same
signal driven from a TTL gate shows 25% overshoot.

So with a bit of a decent driver and termination, the OP can get good
results with simple single ended terminated wire and a 50 Ohm driver. RS-485
or CAN would be overkill.

> I would still not recommend SPI or I2C for this.  It is always better to
> used something intended for the use that you have and neither of these
does.

And when is something intended for the use? IMO anything that gives a good
reliable result.

Meindert

Meindert Sprang wrote:
> "Not Really Me" <scott@validatedQWERTYsoftware.XYZZY.com> wrote in
> message news:7cral9F28uaiiU1@mid.individual.net...
>> Don't try this with SPI, the distance is too great.  I think it is
>> unlikely to work over I2C for the same reason, but I have less
>> direct experience with that.
>
> As Vladimir already pointed out: IDE, GPIB and Centronics were also
> single-ended at TTL level and all worked fine over at least 2 ft of
> flatcable. The SPI cable from my programmer is also 1 ft and I have
> never experienced any problems. Looking at the signals with my
> oscilloscope does not show any ringing, overshoot or reflection that
> matters.
>
> Meindert

Well, he said 4 ft, so that would be more trouble than 1ft.  I did have 
either a 12 or 20 ft Centronics cable, but that is a slow interface.

What speed SPI?  We used ADCs ADuC8xx series and had horrible results at 1 
ft until we used twisted pairs.  It's been a few years but I think it was 
set for 10Mbps but was still flaky at 250Kb, the lowest speed it had.  Only 
choices were 1M, 10M and 250K.

I would still not recommend SPI or I2C for this.  It is always better to 
used something intended for the use that you have and neither of these does.

Scott

On Fri, 24 Jul 2009 14:41:15 -0500, Vladimir Vassilevsky
<nospam@nowhere.com> wrote:

>
>
>bob wrote:
>> On Thu, 23 Jul 2009 22:49:22 +0300, Paul Keinanen <keinanen@sci.fi>
>> wrote:
>
>> So, I modified the receiver for abs(3V) hysteresis. So, the received
>> signal must go higher than 1.5 V and less than -1.5V to change the
>> received state, if sitting at 0.0 V ground.  This should help noise
>> rejection somewhat as well as be able to have some ground differntial
>> Voltage between units of at least a few volts.
>> 
>> So far this method is working great, and should pretty much keep the
>> RS232 compatibility.   Comments are very welcome on this.
>> 
>> I am also using modbus so have a CRC check at the end of the packets
>> and the data rate will not be required to be above, say, 56K baud.
>> Normally, 19.2 K Baud.
>
>The hysteresis won't buy you anything but the problems with the receiver 
>occasionally stuck in the "break" state when the cable is disconnected.


OK, why would the increased hysteresis NOT buy me anything, as far as
noise riding on the signal ?   Say, a 1V P-P noise level ? 

Also, say there was a 1 V dc ground difference between the transmitter
and receiver.

boB




>
>What could be useful is the simplest RC lowpass filter upfront the 
>receiver. Set the RC cutoff frequency at about 1/2 of the bit rate.
>
>
>
>Vladimir Vassilevsky
>DSP and Mixed Signal Design Consultant
>http://www.abvolt.com

On 24 Jul 2009 14:28:01 -0500, bob <bob@bob.bob> wrote:

> Looks like alll RS-232 transmitters transmit + and -
>voltage levels, but hardly any of them xmit higher than +- 6 Volts
>these days.  

The RS-232 standard was designed in the 1960&#4294967295;s to connect
electrotechnical Teletypes (and early VDUs) to a modem and transfer it
over existing telephone network to a central computer along with local
devices. 

The modem was usually in the adjacent rack from the computer or
Teletype, so the 15 m maximum distance was more than adequate and 20
kbit/s was more than even any baseband modem could carry over
telephone wires.

Typical current loop constant current sources had an open circuit
voltage of 24-60 V and telephone circuit voltages are in the order of
40-60 V range. No wonder that the allowed voltages were so high. The
minimum voltage levels are so high, that you could even drive directly
some reed relays :-).

The power dissipation was not an issue with the low speed, short low
capacitance cable connections and since discrete semiconductors and
resistors were used, it was easy to dissipate any power losses.

The situation is quite different today, quite long cables are often
used with a high cable capacitance. At each output transition, the
cable capacitance must be charged or discharged through the
transmitter to the corresponding supply voltage, in which some energy
is dissipated to heat in the transmitter. 

With the current much higher line speeds used, these transitions occur
much more frequently charging and discharging the line capacitance
more frequently and hence dissipating a much larger average power in
the transmitter. With integrated circuits and even with multiple
transmitters on a same chip,  the power dissipation would simply kill
the chip if driving high capacitive loads at high speed using a high
supply voltage.

Looking at the situation at the receiver, the important thing is that
the transition between -3 V to +3 V at the receiver is fast enough,
compared to the line speed used. The transmitter is required to
provide sufficient current in this voltage range to charge or
discharge the distributed line capacitance. 

The problem with the cable distributed capacitance and distributed
resistance is that the initial transmitter output voltage swing must
be larger than +/-3V in order to achieve the +/-3 V voltage swing at
the receiver in a realistic (non-infinite) time. 

Once the voltage at the receiver is outside +/-3 V, there is not much
point in feeding large amounts of current into the line capacitance,
so it is sufficient to feed the very small current consumed by the
high input impedance receiver in the steady state.  

Clearly, the transmitter supply voltage must be larger than +/-3 V
even assuming zero voltage loss switching to charge the cable
distributed capacitance through the cable distributed resistance to
achieve the +/-3 V swing at the receiver in reasonable time (say t=RC
time constant).

The +/-6V transmitter voltage swing is a good compromise, allowing at
least 20 kbit/s into a low capacitance (<15 m) cable (as required by
the RS-232C standard) with a reasonable power dissipation.

Paul

On Fri, 24 Jul 2009 22:31:55 +0200, Heinz-J&#4294967295;rgen Oertel
<hj.oertel@t-online.de> wrote:

>Paul Keinanen wrote:
>
>> For master applications (such as in CanOpen master) or when high
>> (interrupt) latency hardware/OS systems are used, the SJA1000 in
>> PeliCan mode with 64 message Rx FIFO is very handy.
>> 
>
>64 CAN message FIFO is on my wish list for the next generation stand alone 
>CAN controller.

Yes, you are right, the RxFIFO on the SJA1000 is only 64 bytes, so it
could hold only 5-16 standard format messages. 

Anyway, this allows up to 0.5 ms interrupt latencies at 1 Mbit/s or 2
ms at 250 kbit/s.

Paul

On Fri, 24 Jul 2009 14:41:15 -0500, Vladimir Vassilevsky
<nospam@nowhere.com> wrote:

>
>
>bob wrote:
>> On Thu, 23 Jul 2009 22:49:22 +0300, Paul Keinanen <keinanen@sci.fi>
>> wrote:
>
>> So, I modified the receiver for abs(3V) hysteresis. So, the received
>> signal must go higher than 1.5 V and less than -1.5V to change the
>> received state, if sitting at 0.0 V ground.  This should help noise
>> rejection somewhat as well as be able to have some ground differntial
>> Voltage between units of at least a few volts.
>> 
>> So far this method is working great, and should pretty much keep the
>> RS232 compatibility.   Comments are very welcome on this.
>> 
>> I am also using modbus so have a CRC check at the end of the packets
>> and the data rate will not be required to be above, say, 56K baud.
>> Normally, 19.2 K Baud.
>
>The hysteresis won't buy you anything but the problems with the receiver 
>occasionally stuck in the "break" state when the cable is disconnected.

This can occur only if you disconnect the cable during active data
transfer during the space (0) bit. Disconnecting the cable while the
transmitter is in the idle (Mark) space does not cause this problem.

One way around is to use a pulldown resistor at the input to the
negative receiver supply voltage, which takes care of the cable
disconnect problem. However, moving the hysteresis levels to say +0.5
V and +3.0V will also solve any problems associated with powered down
transmitters.

>What could be useful is the simplest RC lowpass filter upfront the 
>receiver. Set the RC cutoff frequency at about 1/2 of the bit rate.

This may be usable in synchronous communication, in which each PLLs
are used to synchronize the receiver clock at the start of connection
or at least long preamble are used in front of a message frame.

However, asynchronous communication is quite picky about detecting the
leading edge of the start bit, so the signal should have a decent
slave rate across the transition area. Thus, I would move the cutoff
frequency at least 1-2 octaves higher.

Paul