S9 Skew investigation - your help needed!

Hi Everyone,

To try and move the Skew investigation forward further, can you check your Flow graphs and make sure they do match up to the Events. e.g. in the graph below you should see that the Obstructive, Central or Hypopnea markings in the Events graph should be at the end or just after the reduced air flow in the flow graph.



If you notice a Skew where the Apnea/Central/Hypopnea don't match up with the flow graph, can you
1. Check your card in the DATALOG folder
2. Look for 2+mb file ending in BRP.EDF (e.g. 20100313_221104_BRP.EDF), the file name is made up as follows:-

20100313_221104_BRP.EDF
YYYYMMDD_HHMMSS_BRP.EDF

Where YY = Year, MM=Month, DD=Day, HH=Hour, MM (second MM)=Minute, SS=Seconds. The BRP is the High Resolution data file. The time within the file and also in the file name should correlate with each other, this is the time you started therapy for the night.



4. Open up Notepad on your PC
5. Drag the file, e.g. 20100313_221104_BRP.EDF onto the Notepad window
6. Verify that the time stamp in the file within Notepad (highlighted in yellow in the left of the image above) is close to the timing in the file name (e.g. within a couple of seconds).

II believe if the timing is off by 6+ seconds then it suggests you may have some skewing going on from what I can tell.

7. Close Notepad (don't save any changes if asked to do so as this will corrupt the data file.

What I'm trying to find out is
1) If you do have skewing but don't see an offset here on the time
2) If the flow graph appears to be fine but you're seeing skews with other graphs like Flow Limitation

For the most part I haven't seen any skewing when power cycling the S9 so I think this may be the workaround to keeping it skew free.

I'd welcome any comments on the subject or any experience you have with skewing you may be seeing so we can narrow this one down further.

Thanks
Dave

I have never seen any skewing of data (23 days), I've looked for it on every day and I've power cycled the S9 every day from the start.

-Corey

Dave,
Apologies for my confusion. There is a "skew" discussion going on and also an "hour lost" discussion for the S9. After 14 days of intense scrutiny (no skew, no gap), I now find an hour (exactly) gap in the detailed data. The summary data looks accurate.

I also noted that on the "missing hour" night (last night) the events prior to the gap make no sense. It shows normal breathing, but an event is marked (twice in this case). The events after the gap are normal.

I believe I have a "gap" problem and not a "skew" problem, but maybe they are related and perhaps you can add some perspective.

Thanks

I don't think you'll see it if you're power cycling the S9 regularly, certainly I am not seeing it when I power cycle it, it seems to be more when you don't power cycle it regularly that it seems to run away.

Hi Dave, I have more questions than answers, I'm afraid.

I went to the Datalog Folder, which I saved before formatting my SD card yesterday. When I look at the files, which start March 15, the 1st date that any BRP.EDF files appear is March 28. I don't know if that's significant or not?

When I drag the BRP.EDF over to Notepad, this is what I see:



The files for March 18 have separate groups of files, ending at 12:50 am, 5:29 am, and 7:12. I go to bed around 10 to 1030 pm. From my notebook, I see that I did get up for a bathroom break at 1 am and turned off cpap. Then again at 539 to adjust bed covers, and woke up at 718 am.

My bedside clock (it's a cheap battery operated one) tends to run fast, so the times probably correct on the data files.

Here's the Detailed Graph for Flow, where you can see the wake-up times:



Something I've never figured out, with ResScan 3.7 and 3.10, is why the times are correct on the Detailed Report, but skewed on the Detailed Graph:



Does anyone else have this problem?

So, back to your questions Dave--if you can tell me how to unscramble the data in Notepad, I'll continue the detective work for you.

I formatted my card on Friday, no skew on Saturday, 1 second skew on Sunday, formatted card on Sunday no skew today. The only difference in the days is that Sunday I took the card out before I turned off the machine. I am turning the machine off each morning and back on each night as much for power savings as for the skew problem. I didn't copy files from the card before the 2 formats but if I see a skew I will try to remember to look at the data before formatting the card.

For those looking at data, my skews are pretty small and aren't visible unless I select 1 minute or 30 seconds for the amount of data to show on the graph at one time. I rarely looked at the data finer than 10 minutes so I didn't see the skew.

Wow - talk about crowd sourcing. This is incredible, dave21. Count me in.

I'm seeing data skew double every day from the data of card wipe - starting with one second the first day. However, data skew seems limited to the beginning and end of line-graph data. I have not let it go long enough that the skew is obvious for the events. It's harder to be sure there is data skew with the events, since they are point data rather than line data.

Example dates from .edf files and file mod dates:

20100405_041127_BRP.edf
05.04.1004.11.28768
4/5/2010 7:41 AM

20100402_215805_BRP.edf
02.04.1021.58.22768
4/3/2010 1:53 AM

Apparently, locking the card, or not, has no effect best practices.
Other sources of variability to consider:
Line amperage/voltage? Seems unlikely.
Computer speed? Seems more likely to be specific to the S9 and not the computer reading the SD card.
Other ideas?

KatieW wrote:Hi Dave, I have more questions than answers, I'm afraid.

I went to the Datalog Folder, which I saved before formatting my SD card yesterday. When I look at the files, which start March 15, the 1st date that any BRP.EDF files appear is March 28. I don't know if that's significant or not?

Hi Katie, that would be correct, because this is a Hi-Resolution file, so on the SD card you'll only have the last 8 days of High Resolution data.

KatieW wrote:When I drag the BRP.EDF over to Notepad, this is what I see:

Hmmm, that doesn't look right, it's definitely the BRP.EDF file and not PLD.EDF that you've dragged into Notepad? If it's the BRP.EDF then maybe your Notepad isn't reading the header correctly, all of my BRP.EDF files have the header information in ascii so you can read the start date and time.

Mine is around 2mb in size but I don't have any breaks through the night so that might maybe affecting the the result you're seeing. Can you try the other larger files >200kb and see if you can read any of those for the same day?

KatieW wrote:Something I've never figured out, with ResScan 3.7 and 3.10, is why the times are correct on the Detailed Report, but skewed on the Detailed Graph:



Does anyone else have this problem?

I haven't seen this skew, I'll go and check my machine when I get back home and take a look to see if I see this.

KatieW wrote:So, back to your questions Dave--if you can tell me how to unscramble the data in Notepad, I'll continue the detective work for you.

You could try to use another freebie editor like TextPad or UltraEdit and see if you get the same results there. I'm wondering if it's either Notepad on your machine isn't reading the ascii part, or maybe it's corrupted.

Thanks
Dave

DreamDiver wrote:Wow - talk about crowd sourcing. This is incredible, dave21. Count me in.

Thanks DreamDiver! I'm really hoping we can get somewhere further to work out what's going wrong and the very least get this back to ResMed to fix in later S9's if it's the S9 firmware at fault.

DreamDiver wrote:I'm seeing data skew double every day from the data of card wipe - starting with one second the first day. However, data skew seems limited to the beginning and end of line-graph data. I have not let it go long enough that the skew is obvious for the events. It's harder to be sure there is data skew with the events, since they are point data rather than line data.

I think this is similar to what I've seen, when it starts to skew it almost doubles and gets compoundly worse and then to the point it skews too much that over time you then lose an hour of flow data.

DreamDiver wrote:20100405_041127_BRP.edf
05.04.1004.11.28768
4/5/2010 7:41 AM

So if I'm right on this, there's a 1.7 second skew (which is probably within tolerable levels)

DreamDiver wrote:20100402_215805_BRP.edf
02.04.1021.58.22768
4/3/2010 1:53 AM

Again if I'm right on this, there's a possible 17 second skew? Do you see see a similar skew in the flow data?

The problem I wonder why some aren't seeing the problem is whether they're not going down to the 30 second graph or higher as that's where it's more apparent as you, Katy and I have been seeing. Once it gets really out of hand you then seem to lose a 1hr block of Flow data towards the end of the night.

When I've seen these numbers not tally up >2 seconds, it seems that they never become in-sync again and continue to get more out of whack as new days go on, and I think (not 100% sure) that this is somewhat affecting in some way the skew.

The next thing I want to experiment with (I need to make a number of backups) is to write out a number of SD cards where the data is skewed and start hex editing these files and only replacing time values and see by doing this if the data comes back without a skew when re-importing the modified data into ResScan. I'm hoping it might just be that easy, but my guess is it probably won't be.

DreamDiver wrote:Apparently, locking the card, or not, has no effect best practices.
Other sources of variability to consider:
Line amperage/voltage? Seems unlikely.
Computer speed? Seems more likely to be specific to the S9 and not the computer reading the SD card.
Other ideas?

Yeah I'm pretty sure it has nothing to do with the computer or not write protecting the SD card when inserting it into a computer, although I still think it's best practise to write protect it so Antivirus software doesn't mess up the files when it tries to scan them.

I'm still sold on the S9 being the corruptor of the data, and if I can find a way to fix the skew temporarily by editing the data then I think that gives concrete evidence to what needs to be fixed permanently to stop it from happening. My guess is there might be a simple rounding exercise in the S9 that isn't rounding properly, e.g. summary data gets rounded down but detailed or hi resolution data gets rounded up. Some nights it doesn't make much of a difference, but it's possible the more start/stops you have throughout a day might be impacting it more. Hence why you and Katie have seen it more than I have.

EDIT:
The only other variable I can think of is are you pre-heating your water? And if so do you do it for the same length of time? I don't think it will be this but it's another thing to perhaps discount from the equation. I could possibly see something going wrong in the software because it knows it's pre-heating the water and it might start the logging earlier some nights depending on the length of pre-heat time is used in some files but not in others. e.g. because the blower is functional during that time frame of pre-heat and also cool-down.

Thanks
Dave

I tried a couple more things to see if I could force a skew or loss of data to happen.

1) set the CPAP to pre-heat for 3 minutes (no 3 minute skew was found this morning on my data)
2) Powered on/off the S9 3x times in the night as I was heavily congested and had to get up to try and clear my nose

In both cases, I didn't see a skew at all this morning or any data loss. I had power cycled the machine before the start of the first session. The difference in the BRP.EDF file was only 2 seconds difference still.

Thanks
Dave

I have a 1 second skew today.

time from the file is 05.01.59768

while the time in the file name is 050158

I pre-heated the humidifier and got distracted so it had started cooling before I started using the machine.

It might be interesting to look at the file using the free edf reader from http://www.edfplus.info/downloads/downloads.html.

1-2 second different in the time stamp I believe is within tolerable limits as it mostly seems this way and mine don't increase past this if I'm following the previous steps I've listed on how to keep the skew from happening.

Thanks for the link, I remembered there was a thread earlier last month on the open source software and I couldn't remember what the link was, I'll check it out to see if I can work this out any further.

Thanks
Dave

Further information found...

Monday Night - Pre Skew
- Prior nights data (Sunday) was not skewed at all
- Power cycled S9 prior to first session
- This was the first night in ages I had a multi-session night (due to congestion)
- Verified no Skew in data the next morning as you can see by the graph below



Tuesday Night - Skew occurred in AHI data
- Prior nights data (Monday) was not skewed at all (as mentioned above)
- Did NOT power cycle S9
- Had a single session usage for the day/night
- Verified Stats Wednesday morning and am now seeing a Skew in the AHI graph
- I was awoken by my alarm at 6:30am and powered off the machine at this point
- I immediately took the H5i reservoir and SD card out of the machine immediately and emptied the H5i Reservoir, then took the SD card to Laptop to read stats
- There is no way that the S9 could have written another 30 mins of AHI data as the card was in my hand/laptop at this point
- I believe as I've mentioned before that the S9 had skewed the AHI graph data



Analysing the BRP.EDF file
- Consistently I see a 1-2 second difference here but never > 2 seconds between what's logged in the BRP.EDF file and the time stamp in the file name
- Last night the difference went to 3 seconds (albeit this is a minor skew in the AHI data)
- I believe if left to continue it will get worse


What did I do differently?
- Last night I did not power cycle the machine before first use
- The previous night I had multi-usage sessions (I'm basing this on the amount of times I was on the machine and powered it off to take a break)

Suggestions
- If you have a multi-session use of the S9, you may find that the data doesn't skew until the day after
- Check the current day vs the next day's data and ensure that all graphs are lining up correctly
- Check for gaps in the Flow Data
- Check that the AI/CA/HI's line up with the Flow graph

DreamDiver, I'd be interested in hearing your thoughts on this seeing that you do multi-usage sessions on the S9. Would you be able to not potentially do multi-sessions for a couple of days and see if you still get a skew?

Thanks
Dave

Dave

I'm not sure I understand you're AHI skew...

The AHI is an Index and reports 0.0 before 3:00am on Wed morning... then it rises between 3 and 4am because of 2 OA's.
It continues at the same rate between 6 and 7am because the Index reports for each hour... so it would report for the whole hour as part of the Index and it always continues to the end of the hour... not as part of your "sleep or flow" and doesn't stop at same time.

I have found that AHI starts at first event and continues past the last event to the end of the hour (as per definition)
I have found that Pressure, Leak, Minute Ventilation and Flow Limitation all report the same timing and turn on and off with S9.
Snore Index starts with Pressure, Leak, Minute Ventilation and Flow Limitation but continues through all sleeps and naps until machine is turned off after last time.
Events (OA, AI, H etc) report separately as occurring.

Nord

Nord wrote:I have found that AHI starts at first event and continues past the last event to the end of the hour (as per definition)

I think you might be right (and I can understand the theory because AHI is by definition an hourly graph, but looking at most of my other AHI graphs these either do stop when the S9 is powered off or before. I'll do some more investigation into this one, but you could be right, this might be a red herring.

Thanks
Dave