Frequency change detection

I am an Orthopaedic surgeon and I came with the following idea.
During a cementless hip replacement, a cromium-cobaltium stem is impacted into the femur. This is done with a few hammer blows. The trick for a successful surgery is to push it as far as it goes without of course breaking the bone.
It is well known amongst senior surgeons that the sound of the hammer blow onto the tool that holds the stem during insertion “changes” and when the ear of a senior surgeon detects this sound “change” he usually stops. Of course this is not the only factor he takes into account to decide whether to stop or not.
Nevertheless I thought it is worth documenting this sound change. I bought a tascam dr-07 recorder and recorded the operation in high quality audio. I tried analyzing the frequency of each hammer blow by selecting the short clip which covers each blow and choosing plot spectrum in Audacity. What I expected was to see a clear trend in the frequency of the sound with the highest db in each consecutive blow. That is because the mass of the stem becomes more “integrated” into the mass of the femur and thus the sound response should be that of a lower frequency sound.
I am a little lost here and clearly far away from my area of expertise.
I attach a sound clip. If anyone has an idea or sees something I am missing any feedback would be highly appreciated.
Here is a photo of the stem itself

First, the answer to the question you didn’t actually ask, Audacity doesn’t do anything in real time except record and play and some very minor timing functions, so no, you can’t use that to tell you about the progression.

Second, impacts are extraordinarily difficult to analyze. Even with tricks, I couldn’t do a definitive job with your sound files. Any of us can analyze a bell and can tell you the pitch with pretty good accuracy, but we can’t do it from the clapper striking the metal surface. We have to do it by analyzing the trailing waveform, the “ring.” That’s the bell mechanically vibrating at a rate that it really likes. Hips don’t ring enough.

I got different analysis results from the changing of the sound, but I didn’t get anything useful. An impact sound is a cousin to white noise (hiss). The analysis of either one is, however briefly, all frequencies. In addition, one impact isn’t enough for the analyze tools to use. I had to duplicate one impact ten to fifteen times to give enough information for the analysis to chew on.

We’ll see what the other elves have to say.


Human hearing is really incredible. The change in sound is very clear, but analysing the sound shows very subtle changes.
This video is rather rough, but hopefully casts some light:

I am happy I got your attention and that I found a place where I can share my enthusiasm and my skepticism.
I can understand the problems of such a project. One of the issues I find troubling is the integration of the impactor with the stem itself. This is the tool which one needs to screw onto the stem in order to impact it. If the screw is tightened more or less this I believe can give very different results.
The annoying thing though is the fact that although the change is audible to the human ear, this seems to be very difficult to quantify.
To be honest there are cases (which I haven’t recorded yet) that the change is even more obvious.
I attach another set of files.
These are from a different surgeon and you can hear the sound of the hip stem and another file from the impaction of the acetabular component.
If anyone is more interested in the form of the metalware this is the technique guide from the company of the hip replacement
The hip impactor is shown on page 8 and the acetabular cup on page 10
Acetabular clip
Hip clip

To be honest, the idea of sawing off the top of the femur, reaming it out and driving in a metal spike make me feel a little queasy, but how the ear is able to pick out the subtle change in sound is fascinating. The same phenomena is encountered in many other situations where a metal spike is being driven into something. I’m English and I drink tea with milk and one sugar and I also find it fascinating how the sound of stirring a cup of tea changes.

Here’s a spectrum from a few hits near the start. I’ve normalised the waveforms so as to produce a similar amplitude scale on these two images. Note how there is relatively little sound in the 1 kHz region.
Now here’s a couple of similar strikes from near the end of the recording. Note how there is a lot more energy in the 1 kHz region due to “ringing”.
On string instruments, the “sustain” is greatly dependent on the rigidity of the suspension points (where the string comes in contact with the instrument - often called the “nut” and the “bridge”). The more rigid the suspension points, the more the string will ring (the greater the “sustain” will be). The same is true of single ended oscillators, such as a steel rule clamped in a vice, or a metal stem driven into a femur.

Well that is the problem with us orthopaedic surgeons. Amongst other doctors our reputation is low (also called orthopods) because our surgeries seem so brutal. Nevertheless they are very technical and they have a lot fascinating details that I personally love :slight_smile:)

Exactly! Same with one spoonful of salt dissolved in one cup of warm water.

Was this in the cup or the stem?
So I will start looking for more output in the 1kHz area in my recordings and not a trend in the peak sound…
Sounds right thought. All the energy in the 8-10 Hz area is from the stem-impactor complex and the 1kHz area the response of the femur (+ all soft tissue around it) when it comes more integrated to the complex.

That’s with the file “forum femoral.wav”
The file “forum acetabular.wav” is more difficult to analyse because the way that he is hitting it changes half way through.

Surely Audacity’s “spectrum” display is a better option to visualize changing frequency content ? …
Audacity spectrum display of hammer blows (contrast increased with GIMP).jpg

IMO when driving a nail into dense wood there are two changes in the sound …

  1. An initial increase in pitch as portion of nail free to vibrate is reduced with each hammer blow as it penetrates the wood.
  2. Reduction in the duration of ringing as the nail is held increasingly firmly the deeper it penetrates the wood, (increased damping).

I think the “that’s enough” sound is when the duration of the ringing falls below a threshold, rather than when the sound reaches a particular pitch, ( the pitch could vary with each operation as would depend on the dimensions of the prosthesis and its composition).

I think the “that’s enough” sound is when the duration of the ringing falls below a threshold

Exactly. Complicated by the actual impact sound which doesn’t change. I wonder about not only amplifying the work, but carefully sheering off the first spike of sound to get rid of the hammer contact. Everything else should be contributed by the actual bone/metal moving.

I’m not in a position to try it, but sheer off the first impact, Amplify, select and duplicate your brains out, Analyze.

I wonder if a lot of that trash at the high end would go away making the character of the sound easier to manage.


But still the impact sound has to progressively change. This “ringing” of the stem has to depend on the mass of the stem. When the mass of the stem becomes more integrated into the femur then logically the frequency of the ringing has to change accordingly. Furthermore how easy is it to separate the two elements of the blow: the hammer hitting the impactor and the stem “hitting” and ringing along with the femur?
Again thanks everybody for the brainstorm.

Could (securely) attach a contact microphone to the prosthesis/patient to isolate its sound from that of the hammer and room sounds, or if that was impractical attach a contact microphone to the hammer and subtract its spectrum from the sound in the operating theatre recorded via a regular microphone.

There are similar engineering problems to yours, e.g. where the sound spectrum of an engine is compared to a reference spectrum to detect the earliest signs of component failure. To do this one spectrum is subtracted from the other (A-B) any difference would indicate an abnormality in the engine.

In your case you’d have to make a spectrum of each hammer blow and have (probably custom made) software compare (subtract) it from the spectrum of the next blow. When there were no changes in the spectra from consecutive hammer blows the prosthesis hasn’t moved, assuming it is being struck in a consistent fashion.

NB: the changes in ringing time wouldn’t show up in an analysis of the spectra though: temporal matters don‘t show on spectral analysis.

I like this idea a lot. Does anybody have a clue how this can be done? Eg where I can find such software or find the person who can build it?

The physics / (electronic) engineering / computing departments of a university.

There may be a ready-made solution: I just Googled this …

M Mulier

University Hospital Louvain, Department of Orthopaedic Surgery, Weligerveld, 1, 3212 Lubeek, Belgium

The operation technique and prosthetic materials for total hip replacement (THR) have continuously improved. Still, defining the end-point of the prosthetic stem insertion into the femur canal relies on the feeling of the orthopaedic surgeon. This consists of a sense of mechanical stability when exerting torque forces on the prosthesis as well as a feeling of the prosthesis being well fixed and not displaceable along the axis of the femur …

The hammering is stopped when the FRF graph does not change noticeably anymore> .

Apparently they clamp on a device which applies mechanical energy (vibration) to the prosthesis in a consistent fashion*, and analyse the frequency content of the resultant ring via sensor on same device, (contact microphone ?).

[* rather than using the hammer blow to cause a ring which could be inconsistent in force and position,
(hitting it at a different angle could cause a different frequency content, due to different mode of vibration)].

This can be easily done and is called spectral subtraction. I have a patch for the noise removal effect to do just that, but I looks like it won’t be accepted in audacity before the 2.0 release.

unfortunately, when (about a year ago) I searched the literature for articles on this idea, I used the wrong keywords.
There are a few papers from a Belgian team of surgeons and bioengineers who tested what we talked about in different ways
They recorded the sound, they used accelerometers in three dimensions, they used vibration sensors on the stem itself and they have published all their results from tests in vitro, in vivo and in cadavers.
You can find some of the articles here if you want some more technical details.
So I don’t think there is anything left for me to test with the equipment I have available and I decided to sell my recorder on ebay , anyone interested you can find it here
Thanks everybody again for all the input.