John Ehlers Toolbox

Hi guys, your forum looks interesting ! I am pretty interested in cycle analysis. In Forex TSD I posted a Goertzel browser with Bartels cycle significance test which improved a lot the precision of the GB composite cycle. In addition I believe you can improve a lot the Hilbert Transform, MESA or Goertzel spectrum analysis by preprocessing an Empirical Mode Decomposition (EMD) which is available in MQL5. The combination with Hilbert Transform leads to the well known Huang Hilbert Transformation which is considered to be a superior method for non-stationary and non-linear spectrum analysis and is outperforming fourier and wavelet analysis at financial time series. But of course you can combine EMD also as first step to Goertzel or MESA analysis. In particular the combination EMD with MESA seems to be promising (here I mean the real EMD like in the MQL5 sample not the simplified of Ehlers). Together with the Bartels cycle significance test we would have a leading edge package. - Now where should we proceed with this discussion, in this thread, another open thread or in your premium section (where I havn't subscribed yet) ?

Hi Boxter, I have not had a chance to try your new Goertzel browser but I will take a look. I did read what you have written about it on TSD but I have not had the time to dig any deeper into it. I have been experimenting with Igors Band Pass Cycle finder and find it useful. I have my own dll for extracting cycle periods but it is limited to just the cycle with the highest amplitude. I do run it on multiple time frames to drive adaptive indicators which I am very happy with. I am always open to discuss new ideas here or in the cycle extraction thread in the premium section.

I will also take a look at the EMD code on the mq5 and see what it looks like. Thanks for posting the link.

Best wishes,

Alex

Limitations of EMD method : Optimized Trend Trading - Page 53 @ Forex Factory

As is the case in wavelet analysis and spectral analysis, end effects are a common problem in most of these methods, although I do agree some can be solved easier than others. In case of EMD, I've read this can be solved numerically by adding extra waves to eliminate the end effect.

There's a second potential problem called 'mode mixing'. From what I have read: 'Mode Mixing' is defined as a single IMF either consisting of signals of widely disparate scales or a signal of similar scale residing in different IMF components. Usually mode mixing is identified by the frequencies of different (usually adjacent) IMFs intersecting each other. To overcome this problem a new noise-assisted data analysis method was proposed, the ensemble EMD (EEMD), which defines the true IMF components as the mean of an ensemble of trials, each consisting of the signal plus a white noise of finite amplitude.
Apparently Wu and Huang have proposed models to deal with mode mixing.

For those interested, I have attached an interesting paper on this.

Cheers, Snow.

Allow me to share one more excellent paper, although it's actually more a manual.

It uses lots of examples and is very well and clearly written.

I was not able to upload it, but ND was kind enough to post it in the next post.

Cheers, Snow

Originally Posted by Snowski

Allow me to share one more excellent paper, although it's actually more a manual.

It uses lots of examples and is very well and clearly written.

Direct Link (I can seem to upload it here, file size might be too large):

http://djj.ee.ntu.edu.tw/Economic%20...ransforms.docx

Cheers, Snow

I uploaded this file (inside zip archive attached)

Originally Posted by newdigital

I uploaded this file (inside zip archive attached)

thanks for the papers !
Attached you find a paper of another EMD analysis using an iterative moving average approach for extracting the IMFs instead of the cubic spline approach used by Huang. This seems to provide more robust EMD results.

Thanks Boxter.

Is that the path you followed in your development of the "EMD Huang" indicator?

Was wondering how you solved the 'end effects' and 'mode mixing' I mentioned above.

Thanks, looks like this is becoming an excellent development path..!

Cheers, Snow.

Originally Posted by Snowski

Thanks Boxter.

Is that the path you followed in your development of the "EMD Huang" indicator?

Was wondering how you solved the 'end effects' and 'mode mixing' I mentioned above.

Thanks, looks like this is becoming an excellent development path..!

Cheers, Snow.

Hi Snowski,
the EMD Huang Indi is still following mainly the Huang style described in MQL5 Introduction to the Empirical Mode Decomposition Method using cubic spline for EMD extraction. I havn't implemented the algo of other variations yet. Nevertheless in my eyes it is sufficient first to work with the Huang EMD for frequency analysis and forecast. Further EMD optimization can be still done.

Here is now the EMD Huang indi. Maybe somebody can extend it to a complete Huang Hilbert Transformation. Igorad had already implemented the Hilbert transformation. The attached mql4/mql5 Huang EMD indi is showing you either single IMFs or combined IMFs using the libs from above mentioned MQL5 Introduction to EMD Method and MQL5 Programming Basics: Arrays. The indi is running well on the new MQL4 platform. It can be considered as a really interesting Bandpass filter. This indi could be used now for Goertzel, Hilbert Transform or MESA frequency analysis or simply for detrending or trend extraction.

You can select between showing single Intrinsic Mode Functions (Single IMF) or compound IMF functions (Bandpass IMF). Total analyzed IMFs are shown as comment in the indi window. Wenn selecting Bandpass IMF you can select the amount of high and low IMF functions which shall be neglected for filtering noise and/or trend.

The attached Class libraries must be placed in the include folder.

Edit - editing not working without a duplicate post