Noise in Baseline in TCD Signal

I am facing problem with my GC 7890B, equipped with TCD and FID. On TCD channel I have 2 packed columns;

  1. Silica Gel
  2. Molecular Sieve

The GC is being used for the determination of permanent gases (H2, Ar+O2, N2, CH4, CO, & CO2) in gas samples. There is occasional noise in the signal on TCD channel. This noise disturbs the baseline correction modes and especially, when the concentration of components is very low, like we are facing problem for H2 in concentration of around 0.20%. The screenshots of the problem being faced are given below;

Here baseline correction mode is set to 'Advance' and it reconstructs the baseline due to noise in the signal and adds additional area which is not part of the component signal and reads it as signal. The results increase from 0.20% to 0.24%.

Here is the same signal and when the baseline correction mode is set to 'Classic' the baseline reconstructs and ignores the extra area which is due to the noise of the signal it read sample as 0.20%.

  1. This issue of noise in signal creates problem when it appears at the start or at the end of the peak.

  2. The location or RT of the signal noise is not fixed, meaning it can appear anytime on any chromatogram, but it creates significant error in low concentration components.

  3. I want to use the baseline correction of 'advance' as it offers other benefit over the other one (classical).

  4. This noise is visible only when you magnify the signal and error is indicated by the higher result, e.., 0.24% as in routine it is 0.20%

What are the possible ways to overcome this problem?

  • Hello,

    Since the spikes are random, I do not think you will be able to resolve the issue with automatic integration actions. You should be able to use manual integration to correct any effected injections. You may want to create a post in the GC forum, or I can move this post to see if those users have any suggestions on the spikes in the TCD signal channel. This could be a column, GC maintenance, or TCD electronics problem causing the random signal spiking. 


  • Looks like the Hydrogen peak is the narrowest one shown during the first 3 minutes of the chromatogram. You should be able to see the calculated peak width of this peak on your data output or report, along with the peak area.

    The Slope Sensitivity interacts with the Peak Width setting, so maybe changing the Slope Sensitivity along with the Peak Width (or Fixed Peak Width) will keep the baseline from detecting these steep spikes while still detecting the gradual rise of the actual Hydrogen peak.

    It appears the Classical baseline is ignoring the noise properly, this is the way the Peak Width integration setting is supposed to function since the noise is so much narrower than a real peak at that RT.

    However with Advanced baseline the noise appears to be also filtered similarly, but the baseline appears to respond to penetrations below the real baseline. This is just the opposite of the intended behavior that is documented in the Chemstation Help.

    Advanced is supposed to ignore penetrations, and Classical is supposed to follow the baseline penetrations for instance in cases where that is the desired approach. There is also the third choice "No Penetrations" which might be worth trying, but it is unclear if that functions as Classical or Advanced. That may be able to inhibit the baseline from following these penetrations. You really just want to ignore penetration here.  The positive noise excursions of the baseline are already being ignored.  You wouldn't want that to be treated as false peaks.

    Sometimes noise like this on a TCD can be from someone shutting the door to the laboratory or turning on a fume hood or other "cabin pressure" upsets. Also from vibrations like bumping into the GC when it is running, or even placing bottles or vials on top of the GC or nearby on the bench. Even heavy trucks traveling nearby can shake the foundation of the lab floor enough to see a little bit of "seismic activity" on the baseline at high sensitivity.

  • Hello Marty, I would appreciate if you will move this post to GC forum. Furthermore, we have changed the column but nothing got any better in baseline spiking behavior.

  • Hello Mike, Thanks for the detailed reply on the issue.
    You are absolutely right that baseline correction modes are behaving opposite to their intended behavior documented in literature by Agilent. At first, my conclusion was also the same but after thoroughly studying the problem I have concluded that in documentation the behaviors of baseline correction modes are supposed to affect the "actual signal", not the noise. This is not the actual signal this is noise spiking. So, I have concluded that there is nothing wrong with the baseline correction modes, this problem is just beyond the scope of the baseline correction mode.
    Secondly, I want to get your opinion on "data rate", if I set the data rate for data acquisition to less sensitive mode, like if now it is of 20Hz and I may set it to 5Hz, then is there a possibility that it may cancel out the noise peaks by averaging out the signal on less sensitive data rate setting?

  • The TCD has a single filament with a valve used to switch between analyzing the sample and reference.  That switching valve oscillates at 5-6 Hz, so using a higher data rate than that can cause more noise.

  • Good Call.

    This is something I hadn't thought about and it does make perfect sense.

    Haven't used the HP-style TCD since the HP5880 in the early 1980's and the switched carrier & reference flow was one of the elegant solutions not shared by other vendors.

    It was great for Natural Gas and LPG work but that was with packed columns where 5Hz would be fast enough data collection rate for just about anything.

    When I was pioneering the use of emerging capillaries for high-purity petrochemicals by FID the early peaks were so narrow I used the fastest data collection rate I could get from the integrators I had.

    Right now I am seeing discrepancy between 20Hz (too slow) and 50Hz (apparently more realistic but needs further investigation) for the early peaks on the fast hydrogen carrier methods.

    Fortunately we've had plenty of memory to handle the bigger raw data file size, unlike then when we were dealing with KB rather than MB!

Was this helpful?