When a Leak Isn't

[PaytonA]

I initiated a previous thread concerning the difference between my maximum leak rate on Rescan versus Sleepyhead. Since then I have been watching this and have found that it occurs all of the time. It is just the statistics. The graphs look identical. I have also noted that the 95% leak rate suffers from the same difference.

I have not methodically scrutinized the respective leak rate graphs for comparison but they look identical. So why the difference in statistics? I have an idea, which is what I want to present.

First the leak rate is calculated based on an offset of the exhalation curve from the expected curve. In order to provide the expectation the software uses some number of curves to set up a zero line and then compares the actual measured curves to the zero line. The offset is the unintentional leak rate.

As a result, it seems to me, that if one is breathing very regularly and then suddenly takes a deep breath that will show up as a spike in the leak rate graph. I imagine that I am simplifying this somewhat but this train of thought made me curious. I started looking at the spikes in my leak rate chart and noticed that for many of these spikes there was also simultaneously a spike in my flow chart.

I also did a small test with my CPAP. After I had awakened in the morning, I stayed in bed for an extra 20 minutes and breathed with different tidal volume and timing. At the same time, I was carefully checking for unintentional leakage all around the mask. I can usually feel leakage but I also checked for leakage using my hands. There was no leakage. When looking at my leakage graphs, there were leaks recorded through that whole 20 minutes. Not large leaks but spiky and higher than previous leakage rates. Made me feel like I was definitely on the right track

Here is what I think may be going on with the statistics. I think that Resmed recognized this and set up an algorithm that ignores these for statistical purposes. Therefore the Rescan 95% and maximum leak rate statistics are always lower than the equivalent Sleepyhead statistics. The median may suffer from the same difference but my median is always 0 which makes it a little difficult to tell.

If anyone has opinions on this situation, please bring them up.

Best Regards,

PaytonA

[worn_out_in_lebanon]

I'm curious about how it rates the leaks. The algorithm could compare inhale and exhale and try to figure it out that way. Or it might compare what the flow rate should be for a given "commanded flow" (voltage to the blower) vs. what the sensor detects the actual flow to be. (greater flow than what it should be for a restricted vent would indicate a leak - because all flow even after inhale and exhale has to eventually be exhausted) Averaged over some period of time.

For example, the computer in my car(s) has a commanded afr. (air/fuel ratio)
I can monitor the actual afr using a wideband o2 sensor.
(tuning with hptuners software, gm vehicles - Camaro, Vette, Impala)

When the actual afr (measured by the wideband) varies from the commanded afr from the pcm (powertrain control module), I know something needs adjusting. A mapping table, injector offset table, etc. Or I have a leak in the exhaust somewhere which is giving me an inaccurate reading at the sensor as additional oxygen is blended into the exhaust stream.

[PaytonA]

The explanation that I was given was that for every inhale flow curve there is a known and predictable exhale flow curve. The system measures a number of inhale curves and uses the projected exhale curves to come up with a baseline exhale curve. It measures actual exhale curves against the baseline and the difference is leak.

Best Regards,

PaytonA

[justMongo]

It does a DFT, Discrete Fourier transform, on the flow data and finds the coefficient for the ω=0 term.
That's the steady state level.

[worn_out_in_lebanon]

In English that means...?

[PaytonA]

We don't know that Mongo actually speaks English.

[eseedhouse]

Well I sort of know what a Fourier Transform is. If I recall rightly it is a mathematical method of converting a signal in the time domain to one in the frequency domain. Which means devolving a complex signal into a number of sine waves.

In theory you can create any complex wave form by adding together pure sign waves at various frequencies and phase relationship. The tricky part is to find out what collection of sign waves will make the complex wave form you want. That's what the Fourier Transform is for, I think.

[AlanE]

In English that means...?

It does something with the other thing.

[justMongo]

Well I sort of know what a Fourier Transform is. If I recall rightly it is a mathematical method of converting a signal in the time domain to one in the frequency domain. Which means devolving a complex signal into a number of sine waves.

In theory you can create any complex wave form by adding together pure sign waves at various frequencies and phase relationship. The tricky part is to find out what collection of sign waves will make the complex wave form you want. That's what the Fourier Transform is for, I think.

Spot on Ed. The sine waves are harmonics of the fundamental. And there is one term where n=0. It's the steady state (or DC value.)
Performing a DFT on the flow waveform and looking at the coefficient for the n=0 term in the sum of harmonics gives the "zero" for the flow data.

Payton's thread title made me think of something entirely different...

[Walla Walla]

This is a oldie but goodie. Had me rolling.   An all star cast.