Radar-based single target vital sensing
In an embodiment, a method includes: generating a target displacement signal indicative of a movement of a human target based on raw digital data generated by a millimeter-wave radar sensor; and estimating a vital sign of the human target based on the target displacement signal, where generating the target displacement signal includes: generating target in-phase (I) and quadrature (Q) (I/Q) data associated with the human target based on the raw digital data, classifying the target I/Q data as high quality data or as low quality data using a first neural network, when the target I/Q data is classified as low quality data, discarding the target I/Q data, when the target I/Q data is classified as high quality data, performing ellipse fitting on the target I/Q data to generate compensated I/Q data, and generating the target displacement signal based on the compensated I/Q data.
1. A method comprising:
transmitting radar signals using a millimeter-wave radar sensor;
receiving reflected radar signals using the millimeter-wave radar sensor;
generating raw digital data based on the reflected radar signals;
generating a target displacement signal indicative of a movement of a human target based on the raw digital data; and
estimating a vital sign of the human target based on the target displacement signal, wherein generating the target displacement signal comprises:
generating target in-phase (I) and quadrature (Q) (I/Q) data associated with the human target based on the raw digital data,
classifying the target I/Q data as high quality data or as low quality data using a first neural network,
when the target I/Q data is classified as low quality data, discarding the target I/Q data,
when the target I/Q data is classified as high quality data, performing ellipse fitting on the target I/Q data to generate compensated I/Q data, and
generating the target displacement signal based on the compensated I/Q data; and
performing adaptive Sinc filtering to generate a vital sign filtered displacement signal based on the target displacement signal, wherein estimating the vital sign of the human target is based on the vital sign filtered displacement signal.
2. The method of claim 1 , wherein generating the target I/Q data comprises:
generating preliminary I/Q data based on the raw digital data;
high-pass filtering the preliminary I/Q data to generate a high-pass filtered I/Q data using a high-pass cutoff frequency;
estimating a power of the high-pass filtered I/Q data;
when the estimated power is higher than a power threshold, discarding the preliminary I/Q data; and
when the estimated power is lower than the power threshold, low-pass filtering the preliminary I/Q data using a low-pass cutoff frequency to generate the target I/Q data.
3. The method of claim 2 , wherein the high-pass cutoff frequency is equal to the low-pass cutoff frequency.
4. The method of claim 1 , wherein classifying the target I/Q data as low quality data comprises classifying the target I/Q data as random body movement (RBM) data or as intermodulation product (IMP) data.
5. The method of claim 1 , further comprising performing wavelet denoising on the target displacement signal to generate a denoised displacement signal, wherein estimating the vital sign of the human target is based on the denoised displacement signal.
6. The method of claim 1 , wherein performing adaptive Sinc filtering comprises generating M Sinc filter outputs using M Sinc filters based on the target displacement signal and generating the vital sign filtered displacement signal based on one or more of the M Sinc filter outputs, wherein M is a positive integer greater than 1.
7. The method of claim 6 , further comprising filtering the estimated vital sign using a Kalman filter to generate a filtered vital sign.
8. The method of claim 7 , further comprising generating, with the Kalman filter, a vital sign variance associated with the filtered vital sign, wherein generating the vital sign filtered displacement signal is further based on the filtered vital sign and the vital sign variance.
9. The method of claim 6 , wherein the estimated vital sign of the human target is an estimated heartbeat rate of the human target, the method further comprising estimating a respiration rate of the human target based on the target displacement signal.
10. The method of claim 9 , wherein generating the vital sign filtered displacement signal is further based on the respiration rate.
11. The method of claim 10 , wherein generating the vital sign filtered displacement signal is based on a first sub-set of outputs of the M Sinc filter outputs when the respiration rate is below a respiration rate threshold, and is based on a second sub-set of outputs of the M Sinc filter outputs when the respiration rate is above the respiration rate threshold, the first sub-set of outputs being different from the second sub-set of outputs.
12. The method of claim 11 , wherein the first sub-set of outputs corresponds to outputs of first Sinc filters having adjacent bandwidth, wherein the second sub-set of outputs corresponds to outputs of second Sinc filters having adjacent bandwidth, and wherein a low-corner frequency of a collective bandwidth of the first Sinc filters is smaller than a low-corner frequency of a collective bandwidth of the second Sinc filters.
13. The method of claim 10 , further comprising:
determining a signal-to-noise ratio (SNR) of the vital sign filtered displacement signal;
when the SNR is above an SNR threshold, performing a Fourier transform on the vital sign filtered displacement signal to estimate the vital sign of the human target; and
when the SNR is below the SNR threshold, performing peak counting on the vital sign filtered displacement signal to estimate the vital sign of the human target, wherein generating the vital sign filtered displacement signal is further based on the SNR.
14. The method of claim 6 , further comprising determining a frequency range of the vital sign of the human target based on the target displacement signal using a deep neural network (DNN), wherein generating the vital sign filtered displacement signal is further based on the frequency range.
15. The method of claim 6 , further comprising generating the vital sign filtered displacement signal based on a plurality of Sinc filter outputs of the M Sinc filter outputs by concatenating the plurality of Sinc filter outputs.
16. The method of claim 6 , wherein each of the M Sinc filters has the same bandwidth.
17. The method of claim 6 , wherein each of the M Sinc filters has a fixed corner frequency.
18. The method of claim 1 , further comprising displaying the estimated vital sign on a screen.
19. The method of claim 1 , wherein a vital sensing pipeline is used for generating the target displacement signal and estimating the vital sign of the human target, the method further comprising:
determining a number of people within a field-of-view of the millimeter-wave radar sensor;
when the number of people is equal to 0, disabling the vital sensing pipeline; and
when the number of people is equal to 1, enabling the vital sensing pipeline.
20. The method of claim 19 , further comprising:
determining a range of the human target;
when the human target is closer than a predetermined range and the number of people is higher than 1, asserting a low confidence signal indicative of low confidence in the estimated vital sign and enabling the vital sensing pipeline; and
when the human target is higher than the predetermined range and the number of people is higher than 1, disabling the vital sensing pipeline.
21. The method of claim 1 , wherein the first neural network is a SincNet neural network.
22. A radar system comprising:
a millimeter-wave radar sensor comprising:
a transmitter configured to transmit radar signals,
a receiver configured to receive reflected radar signals, and
an analog-to-digital converter (ADC) configured to generate raw digital data based on the reflected radar signals; and
a processing system configured to:
generate target in-phase (I) and quadrature (Q) (I/Q) data associated with a human target based on the raw digital data,
classify the target I/Q data as high quality data or as low quality data using a first neural network,
when the target I/Q data is classified as low quality data, discard the target I/Q data,
when the target I/Q data is classified as high quality data, perform ellipse fitting on the target I/Q data to generate compensated I/Q data,
generate a target displacement signal indicative of a movement of the human target based on the compensated I/Q data,
estimate a vital sign of the human target based on the target displacement signal, and
perform adaptive Sinc filtering to generate a vital sign filtered displacement signal based on the target displacement signal, wherein estimating the vital sign of the human target is based on the vital sign filtered displacement signal.
23. A method comprising:
transmitting radar signals using a millimeter-wave radar sensor;
receiving reflected radar signals using the millimeter-wave radar sensor;
generating raw digital data based on the reflected radar signals;
generating a target displacement signal indicative of a movement of a human target based on the raw digital data; and
estimating a vital sign of the human target based on the target displacement signal, wherein generating the target displacement signal comprises:
generating target in-phase (I) and quadrature (Q) (I/Q) data associated with the human target based on the raw digital data, generating the target I/Q data comprising:
generating preliminary I/Q data based on the raw digital data,
high-pass filtering the preliminary I/Q data to generate a high-pass filtered I/Q data using a high-pass cutoff frequency,
estimating a power of the high-pass filtered I/Q data,
when the estimated power is higher than a power threshold, discarding the preliminary I/Q data, and
when the estimated power is lower than the power threshold, low-pass filtering the preliminary I/Q data using a low-pass cutoff frequency to generate the target I/Q data,
classifying the target I/Q data as high quality data or as low quality data using a first neural network,
when the target I/Q data is classified as low quality data, discarding the target I/Q data,
when the target I/Q data is classified as high quality data, performing ellipse fitting on the target I/Q data to generate compensated I/Q data, and
generating the target displacement signal based on the compensated I/Q data.
24. The method of claim 23 , wherein the high-pass cutoff frequency is equal to the low-pass cutoff frequency.