Author |
Topic: Serial port pin driving an interrupt (Read 6285 times) |
|
manxman
Developer
member is offline
Gender: 
Posts: 21
|
 |
Re: Serial port pin driving an interrupt
« Reply #6 on: Jan 1st, 2013, 10:37am » |
|
Richard,
I do appreciate your involvement.
You might be interested to know the context of this problem. I operate a continuous seismic recording station on the Isle of Man using my own design of instrument which, to date has detected 432 earthquakes (most along the Pacific rim). My plan now is to create an array of seismometers on the Island, the data from which can be correlated to determine the direction of arrival of seismic waves and thus pinpoint an event's location. For this it is essential that the data are timestamped to a relative precision of about 20-30 milliseconds. I have considered various options: radio atomic clocks, mains frequency clocks, stabilised crystal oscillators and GPS. The 1Hz pulse from a GPS module is generally stable to better than a microsecond and linked exactly to UTC (=GMT). This option has therefore been selected as the most cost-effective and suitable, provided that the 1 Hz pulse intervals can be captured and subdivided with sufficient precision. It would be great to show that this can be done using BBC4W, rather than in hardware.
Best wishes for the New Year
Manxman 
|
|
 Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #7 on: Jan 1st, 2013, 4:15pm » |
|
on Jan 1st, 2013, 10:37am, manxman wrote:| For this it is essential that the data are timestamped to a relative precision of about 20-30 milliseconds. |
|
You use the term "essential", and yet you also say "about". All one can say, with confidence, is that a Windows software-based timing system is likely to exhibit a classic bell-shaped statistical distribution, where the larger the timing error the less frequently it will occur.
So the question you have to ask yourself is: what would be the consequences of an occasional timing error much larger than the 20-30 milliseconds range? Would it cause an obviously anomalous reading (which you can therefore ignore) or could it give rise to a false reading that might be indistinguishable from a genuine event?
Edit: Here are a couple of 'scholarly articles', which completely contradict each other! My advice would be to assume that the more pessimistic of the two (Myors) is nearer the mark:
http://link.springer.com/content/pdf/10.3758%2FBF03207727
http://psych.cf.ac.uk/home2/chambers/Chambers_2003_BRMIC.pdf
Richard.
|
| « Last Edit: Jan 1st, 2013, 4:38pm by admin » |
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #8 on: Jan 2nd, 2013, 10:29am » |
|
Ground motion data at each seismometer is being logged at a rate of approximately 20Hz, and each data set will be interpolated, using the GPS-derived timestamps, to a UTC-disciplined synchronous rate of 20Hz. The interpolation process will look for and reject any data items with anomalous spikes in timing. The phase relationship between the instruments' data will then be obtained by cross-correlating timeslices of typically 10 minutes (1200 readings): the result being a set of phase differences within the array from which the bearing of an arriving seismic wavefront can be derived.
Apart from earthquake detection, my primary research interest is investigating the character and mechanism of microseismic noise, which is thought to be generated by weather systems passing over the ocean.
Suppose we have a pair of instruments spaced 10km apart East-West, logging microseismic surface waves from the North Atlantic, travelling with a velocity of 3km/s. Then a 100ms difference in their data phase corresponds to a wave arriving with a bearing of 1.7 degrees from north. Thus, even if the timing error is around 100ms, the bearing resolution towards northern sources in this configuration is acceptable for my purpose. Obviously, the resolution is zero with an E-W array for waves travelling E-W, but this would be the case no matter how good the timing. For this reason I will be deploying at least 5 instruments in a broad geographic pattern to ensure sufficient bearing resolution over the angular swathe of interest.
Regards,
Manxman
|
|
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #9 on: Jan 4th, 2013, 4:51pm » |
|
I thought that I would reply to myself with the results of some timing tests I am carrying out!
The 20Hz, RS232 data stream from a single seismometer is being split into two feeds that go to a pair of PCs, each of which has an identical GPS module. The 1 Pulse Per Second (PPS) output from each module is converted to an RS232 voltage level and connected to Pin 6 (DSR) of the port that receives the seismometer data. A second (USB virtual) serial port on each PC receives NMEA sentences from the GPS module.
My program begins by capturing the UTC (=GMT) start time from an NMEA sentence, then adds a second each time the PPS is detected. This PPS event also starts a Windows millisecond counter, so that the interval between PPS signals is divided, yielding UTC time to a (theoretical) resolution of one millisecond. This process is essentially what I have described in earlier posts.
My program logs pairs of data: UTC time as seconds since midnight (e.g. 45678.123) together with the seismic velocity (e.g. -435). Obviously, each PC's log will contain an identical list of velocities, providing an exact basis for comparing corresponding UTC times.
In my experiments, data have been recorded for several minutes, first with a pair of identical Asus PCs, then with an Asus PC and an HP Brio, both running Windows XP.
Comparing the data sets from the pair of Asus machines the UTC timestamps are all identical to the millisecond. In the second experiment I find that about 80% of the times are identical to the millisecond, with the maximum deviation in the remainder being 3 milliseconds.
So far, so good in terms of progress towards my objective!
Regards,
Manxman
|
|
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #10 on: Jan 5th, 2013, 1:23pm » |
|
on Jan 4th, 2013, 4:51pm, manxman wrote:| Comparing the data sets from the pair of Asus machines the UTC timestamps are all identical to the millisecond. |
|
Have you attempted to provoke the kind of event which is likely to upset the timing? For example, have you arranged for context-switches to take place? Have you performed disk accesses? Have you tried to access a network?
According to this Microsoft article a 'quantum', i.e. the time-slice for which a thread will run until a context switch occurs, is either 3.3 ms (single CPU) or 5 ms (multiple CPU):
http://support.microsoft.com/kb/259025
That would imply that timestamps "identical to the millisecond" would be expected only if no context switches take place at all, which is obviously unrealistic. Even BBC BASIC, running on its own with no other applications loaded, has two or three threads!
There's also a relevant discussion at the link below, where somebody states: "Thus knowing how accurate I could get a time-stamp became important. Knowing that Windows would eat 10 or 20 mSec here and there became my limit of accuracy":
http://discuss.joelonsoftware.com/default.asp?joel.3.66319.13
And another relevant comment at the same place: "As for windows being used in control applications, it may not have any true RT facilities, but with a fast enough processor and a low enough workload, you can get some soft RT performance along the lines of 99% of events happen within their time limit.... It also helps if you put in some effort to trap those 1% of errors well and handle them properly".
Richard.
|
|
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #11 on: Jan 5th, 2013, 4:13pm » |
|
Q: Have you attempted to provoke the kind of event which is likely to upset the timing?
A: Yes. I found that when my program's MODE 7 window is dragged about the screen, the timing can miss a PPS count, leading to a 1s step in timing. Hence, I need to find a way to fix the window position and size. With regard to disk access, the program is continually writing pairs of seismic velocity and timestamp data to disk. None of the array PCs will be on a network and no other programs, such as virus checking, will be running, deliberately to reduce the OS's work and minimise the chance of timing glitches.
Q: That would imply that timestamps "identical to the millisecond" would be expected only if no context switches take place at all, which is obviously unrealistic. Even BBC BASIC, running on its own with no other applications loaded, has two or three threads!
A: All I can report is what I have observed by experiment! So far the timing discrepancies are well within the requirements of my project.
Incidentally, when it comes to a real field deployment there are some other interesting effects to consider. The most important factor is that the local geology and topography at each array station can cause a timing offset that may depend on the arrival direction of the seismic wave. This phenomenon is well known and must be taken into consideration during array processing.
Regards,
Manxman
|
|
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #12 on: Jan 5th, 2013, 9:45pm » |
|
on Jan 5th, 2013, 4:13pm, manxman wrote:| I found that when my program's MODE 7 window is dragged about the screen, the timing can miss a PPS count, leading to a 1s step in timing. Hence, I need to find a way to fix the window position and size. |
|
I really think you're looking at this from the wrong angle! If you have observed that dragging a window can lead to an unacceptable timing error, there are probably many other circumstances which can too. Trying to prevent the one specific situation you have noticed (e.g. by fixing the window) whilst blissfully ignoring the possibility (nay, likelihood) that there are others you haven't yet noticed, is in my opinion reckless.
If there is to be any hope that Windows can reliably meet your timing requirements (and I remain sceptical) you need to find a way to make your program tolerant of dragging the window, not prevent the window being dragged! I would have expected that using an interrupt-based method, as previously discussed, ought to mean that missing (rather than delaying) a Pulse Per Second count is extremely unlikely.
Richard.
|
|
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #13 on: Jan 6th, 2013, 10:05am » |
|
on Jan 5th, 2013, 9:45pm, Richard Russell wrote:| I would have expected that using an interrupt-based method, as previously discussed, ought to mean that missing (rather than delaying) a Pulse Per Second count is extremely unlikely. |
|
Further, how do you currently detect the PPS input? If you poll the DSR pin (using for example SYS "GetCommModemStatus") then the likelihood of missing a pulse entirely is quite high, whereas if you monitor the EV_DSR event (using for example SYS "WaitCommEvent") then the likelihood of missing a pulse is comparatively lower. The code complexity is not significantly different (except that the event method doesn't tell you directly in which direction the DSR input changed!).
In either case, what method do you use for preventing excessive use of CPU time? Normally one would use WAIT or SYS "Sleep" to suspend the CPU, but of course they are both deliberately triggering a task switch, which may be exactly what you don't want! But equally sitting in a tight loop using 100% CPU time isn't acceptable either, so I wonder what you are currently doing.
Richard.
|
|
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #14 on: Jan 6th, 2013, 5:04pm » |
|
My program does indeed run around a tight loop monitoring the status of the DSR pin (using SYS GetCommModemStatus), while also checking the RS232 buffer to check whether there are data to receive. If there are, then the UTC time stamp is created by summing: (UTC start time in seconds + number of PPS ticks + milliseconds since last PPS). This process was described in my previous posts.
I note the concerns you raise. However, as I mentioned before I have no experience in assembly programming and would therefore welcome your help in coding an EV_DSR detector to reduce the chance of missing PPS events, for example when the window is dragged or resized. Alternatively, how do I use the SYS WaitCommEvent function you describe to make my DSR detection more reliable?
My program is currently configured to detect the rising edge of DSR but the alternative of falling edge detection would also be OK. Because the PPS is 500ms wide it is important that the same edge is always used.
Regards,
Manxman
|
| « Last Edit: Jan 6th, 2013, 5:05pm by manxman » |
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #15 on: Jan 6th, 2013, 6:00pm » |
|
on Jan 6th, 2013, 5:04pm, manxman wrote:| My program does indeed run around a tight loop monitoring the status of the DSR pin (using SYS GetCommModemStatus) |
|
But how tight?! Obviously you shouldn't sit in a loop using 100% CPU time for an extended period (on a modern PC that could even risk hardware damage) so my question remains: how do you keep the CPU usage under control without sacrificing the timing performance?
Being (apparently) unprepared to let me see your code doesn't make it any easier for me to help you!
Quote:| I note the concerns you raise. However, as I mentioned before I have no experience in assembly programming and would therefore welcome your help in coding an EV_DSR detector to reduce the chance of missing PPS events |
|
I think you may be conflating (horrible word!) two separate issues here. The first is how to ensure you never miss a PPS event entirely, and the second is how to maximise the accuracy with which you measure the timing of the PPS event.
At this stage I am proposing that you concentrate only on not missing any PPS events, which is what you reported happening when you moved the window. To that end I am suggesting that using WaitCommEvent to detect the PPS transitions might be beneficial: that doesn't involve any assembly language or other specialised techniques.
Only once you've succeeded in reliably detecting every PPS event (assuming that's possible) does it make sense to tackle the secondary issue of how accurately you can time their arrival. That might need interrupts, which in turn might need the use of some assembler code, but we haven't reached that stage yet!
Quote:| how do I use the SYS WaitCommEvent function you describe to make my DSR detection more reliable? |
|
The advantage of using WaitCommEvent is that it will store the fact that a transition on the DTR input has occurred, for you to detect later at your leisure.
Richard.
|
|
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #16 on: Jan 6th, 2013, 7:34pm » |
|
Richard,
I've no problem at all listing my code:
Code:REM Data logging with GPS unit and 1PPS timing
MODE 7
VDU3 : REM Disable printer
OFF : REM Cursor off
INSTALL @lib$+"DATELIB"
SYS "timeBeginPeriod", 1
VS1 = OPENUP "COM1: baud=19200 parity=N data=8 stop=1 xon=off"
REM Flush VS1 buffer
*FX 15,1
PRINTTAB(2,5) CHR$(130); " SSM VELOCITY BEAM"
PRINTTAB(0,6) CHR$(130);"_____________________________"
PROCInitTiming
PROCStartDayFile
REPEAT
REM Check PPS pin
SYS "GetCommModemStatus", @hfile%(VS1), ^modemstatus%
IF modemstatus%=32 AND PinHigh=FALSE THEN
PinHigh=TRUE
SYS "timeGetTime" TO Then%
SSM=(SSM+1) MOD 86400
IF SSM=0 THEN
PROCCloseDayFile
DayNumber%+=1
PROCStartDayFile
ENDIF
ENDIF
IF modemstatus%<>32 AND PinHigh=TRUE THEN PinHigh=FALSE
ENDIF
REM VS1 data collecting part
IF EXT#VS1>0 THEN
SYS "timeGetTime" TO Now%
A$=FNGetStringFromPort(VS1)
SSM$=STR$(SSM+(Now%-Then%)/1000) : Velocity$=FNExtractString(A$,1): Beam$=FNExtractString(A$,2)
Info$=" "
MID$(Info$,1)=SSM$ : MID$(Info$,14)=Velocity$ : MID$(Info$,24)=Beam$
PRINTTAB(0,7) CHR$(130); Info$
PRINT#DayFileHandle,SSM$+" "+Velocity$: BPUT#DayFileHandle,10
ENDIF
UNTIL FALSE
END
DEFPROCStartDayFile
DayFileName$=FN_date$(DayNumber%, "yyy MM dd")+".txt"
REM Replace gaps with "-"
MID$(DayFileName$,5,1)="-"
MID$(DayFileName$,8,1)="-"
PRINTTAB(8,0) CHR$(134); DayFileName$
DayFileHandle=OPENOUT DayFileName$
FileOpen=TRUE
ENDPROC
DEFPROCCloseDayFile
CLOSE#DayFileHandle
FileOpen=FALSE
ENDPROC
DEFPROCInitTiming
REM Initialises the timing using PC clock. Start by waiting for the seconds to change.
Then$=TIME$
REPEAT
Now$=TIME$
UNTIL Now$<>Then$
SSM=3600*VAL(MID$(Now$,17,2)) + 60*VAL(MID$(Now$,20,2)) + VAL(MID$(Now$,23,2))
SYS "timeGetTime" TO Then%
DayNumber%=FN_today
PinHigh=FALSE
ENDPROC
DEF FNExtractString(InStr$,WhichStr%)
REM Extracts sub-string WhichStr% (counting from left) in InStr$. NB: Assumes single spaces between sub-strings
LOCAL StrIndex%, Left$, Work$
Work$=InStr$
IF WhichStr%=1 THEN
Left$=LEFT$(Work$,INSTR(Work$," ")-1)
ELSE
FOR StrIndex%=2 TO WhichStr%
Work$=MID$(Work$, INSTR(Work$," ")+1)
Left$=LEFT$(Work$,INSTR(Work$," ")-1)
NEXT
ENDIF
=Left$
DEF FNDayToString(Offset%)
REM Formats today's date (with offset) to a string of form yyyy-mm-dd
REM Get Julian Day Number
Day%=FN_today+Offset%
REM Format date string
Date$=FN_date$(Day%, "yyy MM dd")
REM Replace gaps with "-"
MID$(Date$,5,1)="-"
MID$(Date$,8,1)="-"
=Date$
DEF FNGetStringFromPort(Where%)
LOCAL Data%, Reply$
Reply$=""
REPEAT
IF EXT#Where%>0 THEN
Data%=BGET#Where%
IF Data%<>10 AND Data%<>13 THEN Reply$=Reply$+CHR$(Data%)
ENDIF
UNTIL Data%=10
=Reply$
In the program "VS1" refers to the seismometer, "Velocity" is the seismic velocity and "Beam" is a status value output from the seismometer. Velocity and Beam are sent as data pairs at about 20Hz from the instrument, followed by <CR><LF>. "SSM" means Seconds Since Midnight. You can see that the program produces a single file for each day's recording.
In the code you will see that the time is initiated by reference to PC's OS time (TIME$). Originally I was using the GPS time code contained in NMEA strings to initiate the UTC timing. The GPS unit communicates via a USB-RS232 virtual driver on "virtual" COM3 whereas the VS1 uses COM1. I found that there seems to be a problem such that when COM3 has been opened and closed once, it cannot be opened again. Thus, I now set the PC clock to UTC time using the NMEATime utility before starting the program I have listed.
My use of SYS timeGetTime follows your suggestion in an earlier post, while SYS GetCommModemStatus is used as per the BB4W documentation. I am doing what I can to the best of my knowledge! Any help you can provide to improve matters would be appreciated.
Regards,
Manxman
|
| « Last Edit: Jan 7th, 2013, 09:15am by manxman » |
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #17 on: Jan 6th, 2013, 10:07pm » |
|
on Jan 6th, 2013, 7:34pm, manxman wrote:| I've no problem at all listing my code |
|
For future reference, please put listings in 'code' tags (that's what they're for!): it will result in the use of a monospaced font and preserve indentation, making the code much easier to read, like this:
Code: REPEAT
REM Check PPS pin
SYS "GetCommModemStatus", @hfile%(VS1), ^modemstatus%
IF modemstatus%=32 AND PinHigh=FALSE THEN
PinHigh=TRUE
SYS "timeGetTime" TO Then%
SSM=(SSM+1) MOD 86400
IF SSM=0 THEN
PROCCloseDayFile
DayNumber%+=1
PROCStartDayFile
ENDIF
ENDIF
IF modemstatus%<>32 AND PinHigh=TRUE THEN PinHigh=FALSE
ENDIF
Preserving indentation is particularly valuable for spotting errors in the program's structure, which evidently there is here because the final ENDIF lines up with the REPEAT! Did you not spot that yourself? It's also evident that you didn't run the Cross Reference utility, since that reports 'Unexpected ENDIF'! Try to make full use of the facilities provided to check the correctness of your code.
I asked how (and whether) you are avoiding using up 100% CPU time; on a quick glance I can't see anything in your program to do that, which worries me. If you haven't already done so, use Task Manager to see what the CPU usage is when your program is 'idling' - i.e. doing nothing except waiting for the next PPS event or the next received serial character. The usage should be close to zero, with nearly all the time spent in the System Idle Process.
As discussed previously the usual ways of releasing the CPU - WAIT or SYS "Sleep" - may not be ideal in your application, given its critical timing requirements. In that case you might be better off using one of the dedicated Windows waiting functions, such as SYS "WaitForSingleObject", which are designed to reactivate the waiting task as soon as an event occurs.
That's potentially another reason to consider using WaitCommEvent, because that would need to be used with WaitForSingleObject, and therefore kill two birds with one stone. So in your position I would want to change from a 'polling' method to one based on events, specifically the EV_DSR event, thereby both slashing the CPU time used and potentially reducing the likelihood of missing a PPS count.
Quote:| I found that there seems to be a problem such that when COM3 has been opened and closed once, it cannot be opened again |
|
Have you contacted the vendor of the GPS unit about that? It would seem to be quite a serious fault, which you may find is fixed in a firmware or driver update.
Richard.
|
| « Last Edit: Jan 6th, 2013, 10:10pm by admin » |
Logged
|
|
|
|
manxman
Developer
member is offline
Gender:
Posts: 21
|
|
Re: Serial port pin driving an interrupt
« Reply #18 on: Jan 7th, 2013, 09:14am » |
|
Q: So in your position I would want to change from a 'polling' method to one based on events, specifically the EV_DSR event, thereby both slashing the CPU time used and potentially reducing the likelihood of missing a PPS count.
A: Once again, please show me how to do this.
Q: Have you contacted the vendor of the GPS unit about that?
A: Yes.
Regards,
Manxman
|
| « Last Edit: Jan 7th, 2013, 09:15am by manxman » |
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #19 on: Jan 7th, 2013, 11:56am » |
|
on Jan 7th, 2013, 09:14am, manxman wrote:| Once again, please show me how to do this. |
|
All the information you could possibly need, and more, can be found at MSDN under the various API functions:
http://msdn.microsoft.com/en-gb/library/windows/desktop/aa363479.aspx
http://msdn.microsoft.com/en-gb/library/windows/desktop/aa363435.aspx
http://msdn.microsoft.com/en-gb/library/windows/desktop/ms687032.aspx
The BB4W structure declarations can be got from API Viewer, and the Windows Constants utility will take care of defining the constants. You can either create the event you need using CreateEvent or, more easily, use the @hevent% System Variable.
It's a technique which is new to me too, but I'm confident it is the right way to go in your application. If you get stuck with the BBC BASIC translation, just ask here again.
Richard.
|
|
Logged
|
|
|
|
admin
Administrator
member is offline
Posts: 1145
|
|
Re: Serial port pin driving an interrupt
« Reply #20 on: Jan 9th, 2013, 1:41pm » |
|
on Jan 7th, 2013, 11:56am, Richard Russell wrote:| It's a technique which is new to me too |
|
Although I've not got any easy way of simulating a PPS input to a modem-status line, I realised that on one of my other PCs I'm receiving a demodulated MSF signal that way:
http://www.rtrussell.co.uk/msf/msf.html
The pulse widths are very variable, but it has allowed me to demonstrate that using WaitCommEvent in conjunction with WaitForSingleObject do indeed allow me to monitor the input transitions (in my case they're on RLSD rather than DSR).
I can't easily tell whether I'm detecting the edge timings any more accurately than I would by simple polling, but I'm pretty confident that waiting in WaitForSingleObject rather than Sleep should give benefits. After all, Sleep is effectively saying to Windows: "I've got nothing to do right now, go off and do something else and return to me at your leisure" whereas WaitForSingleObject is saying "I'm very interested in this event, return control to me as soon after it happens as you can"!
Later: Moving the window around doesn't seem to affect the timing significantly.
Richard.
|
| « Last Edit: Jan 10th, 2013, 08:18am by admin » |
Logged
|
|
|
|
|
