 We have seen that the delay by physics speed of sound and the geometry of the microphones allow us only to measure plus minus 90 degrees. Well, what if this is not enough? What can we do? Well, increase the sampling speed sounds a good idea because in our geometry what we can change is only the speed of sound which we can't change as easily or distance between the microphones. Well, also not very easy to change right now. So the only thing we can change is the sampling frequency and we can do it physically so you just increase the frequency of the microphones or virtually by DSP means to interpolate the signal to higher frequencies. This is also possible. So what if we don't want to use physically because simply the MEMS microphone is not fast enough or you don't want to have such a high rate of codec communication? So for sure we will choose the second option to show the DSP capabilities. So how can we do it? The techniques is called interpolation. Okay, it's pretty easy. You take your two signals which are sampled at 16 kilohertz and you put artificially some number of zeros in between these samples. So it would look like this. In our case, let's put three zeros between every sample. So we actually get four times more samples in the signal. So it would look like this. We have a samples and three zeros and again and again. And as a next step we leave it to pass through the low-pass filter. So as a result out of this signal we get such nice signal which will be exactly the same like before. But virtually with four times higher sampling frequency. So if we use this output of the interpolation as the input of the correlation function, we are able to determine the delay in plus minus four in this case. So by these tricks you can virtually increase the sampling frequency and use it in the same algorithm as before to get much nicer resolution of the delay. So this is implemented in our example as well. You just need to first of all do the interpolation flag set to one and use one of the filters. Okay, I think better is the filter because it's much easier, but it would work with both of those. But I think you need to use only one at a time. So if you do this, just enabling the do interpolation together with one filter and the correlation as well. You should get the resolution of the delay plus minus four. So you need to have three flags enabled now. Filtering interpolation and correlation. So three DSP operation one after another. I've got this delay here live on the screen as well. So if I'm running from the left side I got minus four. On the right side I've got three, four. And then if I move around I got better resolution of the delay. Another example of some DSP operation. Quite easy to run and quite interesting results. You can get out of this. And you see the CPU load is still about 12.5%. So still the correlation takes it's 10%. So much more than we need actually. If you just look what happens in the code. With the interpolation what we do is that here by a simple loop we add some zeros in between the samples. And this is particularly all we need to do for interpolating because the filter we developed in the last part. We use the same filter and the correlation was done in the previous step. So just putting few zeros in between samples you get virtually higher sampling frequency for this application.