I got a short pulse signal , which mean it has a very broad spectrum. While a noise component with a single certain frequency right exists within the signal in frequency domain. In short, noise spectrum and signal spectrum are superimposed. How can I get the optimized signal with less noise?
Note : I got the noise spectrum from the partial data before the short pulse. In this case, I assume that the amplitude of the noise are certain in the superimposed signal in frequency domain. So, if i can get the phase information of noise, I can do difference between superimposed signal and noise in frequency domain. Then, a optimized signal with less noise will be got via inverse fft.
Hi Mingsheng,
Currently, I am working on features identification from a short data burst (1.0 s). In my case, usually, low-pass filter would be helpful for noise reduction. But I don't know the duration/length of your short pulse signal. Maybe you can find the answers from the following works:
Article A Review on Feature Extraction and Noise Reduction Technique
Koli MA. Review of impulse noise reduction techniques. International Journal on Computer Science and Engineering. 2012 Feb 1;4(2):184.
Hope it could help you.
Best wishes.
Kun
Dear Mengsheng
Welcome,
You have a narrow pulse signal with an interference signal with single frequency.
So you can use notch filter to eliminate the noise, The center frequency of the notch is the frequency of the single component noise.
Best wishes
Dear : Mingsheng Li
If you know the signal generator, you can train a FIR or IIR digital filters and after learning phase, you can test your filter to get the noise free signal from noisy input approximately. You can also use neural networks instead of adaptive filter.
Dear Abdelhalim Zekry and Ashraf A.M.KHALAF
Thanks for your answers
A notch filter and FIR & IIR filter will certainly eliminate the noise with single frequency, while it also wipes the partial signal with the same frequency. So in this case, any filter will have this problem.
Best,
Dear Mengsheng,
welcome again,
In order to preserve the whole signal and eliminate the single frequency noise you can regenerate the single frequency by using a phase locked loop and subtract it from the global signal.
This will eliminate the noise without any part of the signal.
Best wishes
If you consider your signal as a step response in the time domain, you should be able to calculate the theoretical frequency response of that signal without noise. If you then consider the noise in narrow-band FFT, it could theoretically be "white noise" (a straight horizontal line) which again can be mathmatically derived, as equal energy at all frequencies of interest.. Then get your DSP to take the noise from your signal in real time in the frequency domain, after transforming your step response.
Its not ideal, and it cannot work perfectly in practice, but you will see a significant noise reduction. Furthermore, if you have a frequency band of interest, you can remove out of band signal and noise by a band pass filter. This will help a little, as out of band noise can interfere slightly with in-band products.
You can look a these theoretical cases in Matlab or similar and add filters using the Signal Processing Toolbox. When you have built the model, then your understanding of the problem will be clearer. Have a good think about the step response in the time domain, and then transform it into the frequency domain. Also look at control theory and Laplace transforms for pulse responses. This provides useful maths that will give you a theorectical siolution. Your practical solution at the end of the day will be able to yield some improvement. I hope you understand what I'm getting at? All the best, Mike.
If you know your signal generator, and give signal conditioning first, and then provide the clean signal to your processing stage you are going to avoid a lot of problems
First try to detect the location of pulse. You can possible use non-linear energy detector. Then you can denoise the signal by windowing the signal around the detected pulse.
Dear Mingsheng Li ,
You can use different type of filters (adaptive filter for example...) .
You can also try the multi-scales product analysis (MP analysis). The MP is designed to detect singularities and enhance signals by eliminating noise.
Best wishes,
If your pulse is repetitive, even in a very low rate, you can try time averaging in order to lower the noise floor, provided that you can trigger your equipment by the pulse. Since the noise is not correlated, its average will tend to zero. This is also valid for spectrum analysis: the noise floor will be lowered if your equipment is able to make the averaging of many spectra. Hope I have helped.
Another way of cleaning up an electrical signal is to pass it through a buffer. A non-inverting CMOS buffer could clean up the rising and falling edges of your signal completely. However, there will be a phase delay from input to output due to the propagation time of the gate. But you may be able to still use it depending on your application? If your signal is not compatible with the input to a gate, you can amplifiy it or attenuate it to suite.Likewise, the output can be amplified or attenuated as required. This could be the best option for cleaning up a signal, if you can live with the propagation delay. Best regards, Mike.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing