W31 - Principles of Filtering Biomechanical Data

Question 1

What is human movement in terms of frequency and amplitude?

Answer 1

Low frequency and high altitude

Question 2

What does frequency and amplitude mean?

Answer 2

Frequency: How often changes happen per cycle
Amplitude: The size of the change per cycle

Question 3

What does differentiating do to the data?

Answer 3

Amplifies the magnitude of the data (increases noise per step)

Question 4

What does differentiating displacement do (and next step)?

Answer 4

Displacement -> velocity -> acceleration

Question 5

Why do we differentiate the data?

Answer 5

Want to remove as much error as possible (anything high frequency that isn’t human movement)

Question 6

What are the different types of data filters?

Answer 6

Moving average (3 point)
Polynomials
Digital (Butterworth, most common)

Question 7

Pros and cons of using moving average (and what is it

Answer 7

Raw data is used to calculate average of a point either side of the point of interest

Very simple to use

Doesn’t distinguish between signal and noise
Was only really used when computers couldn’t do complicated stuff

Question 8

Pros and cons of using polynomials (and what is it

Answer 8

Variables and coefficients at different orders (larger order = more complex movements)

Good for projectiles etc..
Helps when calculating velocity and acceleration

Don’t like rapid changes
Struggles to accurately model high frequency inflections

Question 9

What is a spline?

Answer 9

A number of overlapping polynomials in a sequence

Better at dealing with high frequency signals than polynomials

Sensitive to knots (points of overlap)

Question 10

What is a Butterworth filter?

Answer 10

Uses raw and and processed data to maintain the signal but reduce the noise
Cuts off data at a certain frequency)

Question 11

What are the different types of Butterworth filter (explain them)

Answer 11

Low pass (removes high freq noise)
High pass (removes low freq noise)
Band (lets specific data through in middle section)
Notch (lets specific data through in middle section)

Question 12

What are the risks with higher order data?

Answer 12

More oscillations means risk getting rid of some of the signal, likely to also add some data that doesn’t represent human movement

Question 13

What does lag (phase shift) mean?

Answer 13

Butterworth filtering causes a time delay
Have to run data forwards and backwards to put the shift back to the right place
Zero lag filter doubles o.g order (e.g. 2nd order is 4th order)

Question 14

What do you have to consider when applying a filter?

Answer 14

Frequency of movement
Sampling frequency
Noise frequency
Marker specific filtering 
Application of analysis