Optimized compression for spectrographic data transmission at low bandwidth

Greatly compresses the data while saving power necessary for transmission and retaining parameters of interest

Communications Software & Information Technology

The figure above shows the transformation of time domain data 40 into a spectrogram 42. Time domain data includes the time segments 44. A number 46 of time segments 44 are provided to discrete Fourier transform which produces the spectrogram 42. The spectrogram provides signal frequencies grouped into frequency bins 48, shown as the columns of the spectrogram, versus time, shown as the rows. Shaded pixels, such as pixel 50, show the presence of a signal in that frequency bin at the specific time. The spectrogram is then filtered (Gaussian) to remove noise, then binarized (non-maximum suppression method) for Hough transform processing and then transmitted.

Sensors are all around us collecting data on weather, climate, traffic, air, and water quality, and a host of other parameters too numerous to list. Some of this data is very rich and the collection and transmission are continuous thereby taxing communications networks. Transmission is also power-intensive and can be the most difficult aspect of the power budget of remote sensors. Thus, minimizing transmission time saves sensor battery, providing a longer operational life. A shorter transmission time also avoids needlessly occupying the communications link when the link is bandwidth limited.

Further complicating remote sensor data transmission, many applications send the raw time series data or spectrogram over the link. This is often wasteful since there is extraneous information in the raw data, but the signal processing required to pull salient information may be too computationally intensive for a power-conscious remote sensor.

Of course, data compression techniques have been employed to shrink data, but without any knowledge of the signals of interest, they invariably result in suboptimal compression or cause the loss of important information. Further, the user receiving the data may be unaware that any data is missing.

Addressing the above, Navy scientists have developed a data compression and transmission method that includes receiving sensor data which is digitized and transformed into a spectrogram. The spectrogram is filtered and converted into a binary representation and a Hough transform is used to find lines in the representation. Related lines are combined and then converted back into time-frequency space. Lines are optimized and composed into a binary message, transmitted and received at a remote location where a reconstructed spectrogram can be created from the lines. Transmission can be electrical, optical or acoustic over a wire, fiber optic cable, or radio transmission. Accompanying the transmission are parameters such as power, stability, width, and wander.

Compression by approximating the spectrogram as a collection of lines is a reasonable assumption when the signals of interest contained in the spectrogram are continuous wave (CW) tones or linear frequency modulated (LFM) signals. Each line is assumed to last for the duration of the spectrogram window. By breaking a spectrogram into windows representing a short enough duration for the specific application, an arbitrarily complicated frequency modulated signal can be compressed. Overlapping windows can also be used if necessary to obtain important signal characteristics.

Do you have questions or need more information on a specific technology? Let's talk.

Contact Us