Smoothing

Syntax: vout = SMOOTH(x,y,xout)
vout = SMOOTH(x,y,xout,w)
mout = SPLSMOOTH(x,y,n)
mout = SPLSMOOTH(x,y,n,w)
mout = BIVSMOOTH(x,y,z,mx,my)
Defaults: w[1:#] = 1

Smoothing means that the output may not pass through the original data. If it is important that the curve pass through the original data, use the INTERPOLATE function, unless your independent variable is not monotonically increasing. In that case, use the SPLINTERP function. For 2D interpolation, use the BIVINTERP function.

The SMOOTH function calculates a smooth curve using the method of cubic splines under tension. The data is contained in vectors x, the independent variable, and y, the dependent variable. A vector is returned with the same length as xout, the output locations vector. Vectors x and xout must be strictly monotonically increasing, with x[1] ≤ xout[j] ≤ x[#] for j=1,2,...,LEN(xout)

SPLSMOOTH function

The SPLSMOOTH function calculates a smooth curve through the data contained in x, the independent variable, and y, the dependent variable. x need not be increasing. The number of output locations is given in scalar n. The output of this function is a matrix with n rows and 2 columns. The first column will contain the output locations and the second column the smoothed values.

The points are first parameterized in terms of normalized arc length. The normalized length of x is the real length divided by the range of x, that is, the maximum value minus the minimum value. The arc length at a point is approximated by the sum of the lengths of straight line segments connecting all points up to that point. A spline under tension is calculated for x versus arc length and y versus arc length. The x and y values are interpolated separately and then combined to form the output interpolant.

Weights

If no weights, w, are entered, the weight at each data point defaults to 1. The weights control the amount of smoothing at each data point. As the weight at a point decreases, the spline fits that data point more closely.

Spline tension

Depending on the tension, the smoothed curve may not pass through the original data points. The tension corresponds to the curviness, and must be > 0. As the tension decreases, the amount of smoothing decreases and the data points are fit more exactly. If it is close to zero, each smoothing function is almost a cubic spline and the resulting curve is quite loose. As the tension increases, the fit straightens and has less curvature at peaks, valleys and endpoints. If the tension is large, the smoothing is almost linear and the result approaches the least-squares line through the data. The default value for the tension is zero, and the tension is controlled with the TENSION characteristic.

Method

Given a set of abscissae: , a cubic spline function over the region   is composed of cubic parabolas

where , which join at the endpoints such that , , and are continuous.

The smoothing function is constructed by minimizing

subject to the constraint

where    are the weights and    is the spline tension.

The solution proceeds by the standard methods of minimizing the functional

where   and   are auxiliary parameters. The functional is minimized with respect to   and   by setting the partial derivatives with respect to   and   equal to zero.

2D smoothing