Understand How Your Heart Works
Most of us do not bother about what the internals of our bodies do till we have an issue. Then when the doc says this or that, we scramble to understand what is happening.
All of us have a heart, but do you know what it does, how it functions, and how to measure its health? Most of this information is available on the Net, but I never bothered to read about till there was a requirement. At Sosaley we have developed a Wireless ECG nicknamed heartTarang. When we were reading the data from the sensors, we realized it had a lot of noise. When you do any kind of filtering, you first have to understand the difference between good data and noise. Noise filters are available in abundance as are the algorithms for filtering. But you cannot simply implement them without understanding the rudiments of DSP.
In digital signal processing (DSP), the quantum of noise in a data packet is called Signal To Noise Ratio, or SNR. In DSP for things such as music, a lot can be achieved by simply chopping off frequencies above a threshold. This is what MP3, for example, does. In most cases, all signals above 15kHz are just chopped off in a song that may have data between 10Hz and 25kHz. An unfiltered three-minute song will occupy roughly 40MB of space. When converting to MP3, this is reduced to 3 or 4 MB. In audio circles, there have been innumerable wars fought over this with a clear demarcation between audiophiles and normal human beings.
When reading signals of the heart, we are looking at a much narrower bandwidth over which the signals have any value. This is usually between 0.5Hz to 150Hz. And, needless to say, we cannot chop off anything. Instead of using hardware noise filters that are prone to errors, we developed a software noise filter that works in real-time as the signal comes in.
To understand filtering, we first had to understand how a heart works, what data it creates, and what are the segments of an ECG signal. This is explained by my colleague Dr. Muthukaruppan Gnanadesigan in as simple a form as possible here.