Go to the previous, next section.

# Changing the Resolution of a Scan

It is often necessary to change the resolution of a spectrum or continuum scan, for example by smoothing it to improve its signal-to-noise ratio, or by smoothing to obtain data sets at a common resolution to permit intercomparison. There are a number of ways in which this can be achieved simply in UniPOPS,

```	i)   Boxcar smoothing
ii)  Hanning smoothing
iii) Gaussian smoothing
iv)  Smoothing with an arbitrary user-defined function
v)   Convolution and deconvolution with an arbitrary function
vi)  Cross-correlation permitting optimal-filtering
```

Each of these operations will now be described.

## Boxcar Smoothing

To give boxcar smoothing, the verb BOXCAR averages together an ODD number of consecutive data points centered on each element of Array (0). You should set the adverb NBOX equal to the number of points to be averaged (initial value, NBOX = 3).

Suppose you wish to average five consecutive points to smooth the scan in Array (0), type,

```	>NBOX = 5
>BOXCAR
```

## Hanning Smoothing

Hanning smoothing operates on the data in Array (0) using the algorithm,

```	x(i) = 0.25*x(i-1) + 0.5*x(i) + 0.25*x(i+1)
```

This operation is performed by invoking the verb HANNING, which employs no adverbs. To Hanning smooth the data in Array (0), just type,

```	>HANNING
```

## Gaussian Smoothing

The verb CHNGRES smooths the data in Array (0) from its present resolution to a lower resolution specified by the adverb NEWRES (initial value 0). NEWRES must be specified in channels, and be > 1 although it need not be an integer.

CHNGRES computes the width of the required Gaussian smoothing function from the present resolution (assumed to be the channel separation), and the value in NEWRES, and convolves the spectrum with this smoothing function. Note that NEWRES must exceed the present resolution.

For example, if the data in Array (0) has a channel separation of 500 kHz, and it is required to smooth it to 2.2 MHz, type,

```	>NEWRES = 4.4
>CHNGRES
```

## Smoothing with an Arbitrary Function

Sometimes you may wish to define the function with which the data is to be convolved. This function, which must not exceed 11 elements in total, is placed in the array adverb SMWGT. The number of elements in the function is placed in SMWGT(1), and the values of the function go into all, or part, of SMWGT(2) - SMWGT(12). If the smoothing function has N points,

```	SMWGT(1) = N
SMWGT(2 to N+1) = the smoothing function
```

When SMWGT has been set up, the convolution operation is performed by invoking the verb SMOOTH.

For example, if it is required to convolve the data in Array (0) with the function, 0.125, 0.25, 0.5, 0.25, 0.125, then type,

```	>SMWGT = 5, 0.125, 0.25, 0.5, 0.25, 0.125
>SMOOTH
```

NOTE : If the number of points in the convolving function are odd (as in the above example), then the center of the function lies in SMWGT((N+3)/2). If the number of points is even, then the center of the function is in SMWGT((N+4)/2).

An even more general facility, allowing both general convolution and deconvolution, exists through the verb pair CONVFCN and DECNVLV. CONVFCN computes the convolution of the contents of Arrays (0) and (1). DECNVLV deconvolves the values in Array (1) from those in Array (0). For both CONVFCN and DECNVLV, the number of data points in the two arrays must be identical, and the data in Array (0) is replaced by the results. Note that DECNVLV can be used to obtain the deconvolution of over-sampled data.

Suppose you have two data arrays in Arrays (0) and (1) and want to convolve them together, type,

```	>CONVFCN
```

Following this, the convolution of the two functions now resides in Array (0). If you now want to deconvolve this result and get the original data back in Array (0), type,

```	>DECNVLV
```

## The Cross-Correlation of Arrays and Optimal-Filtering

The verb CROSSFCN computes the cross-correlation function of the contents of Arrays (0) and (1). The data in Array (0) is overwritten with the results of the cross-correlation. The number of data points in the two arrays must be identical.

CROSSFCN can be used to optimally-filter your data. Suppose your observed spectrum (scan number 1000) is rather noisy, but you are sure that a line exists within it, and probably has a line-width of 5 channels. You can optimally-filter the data by constructing a Gaussian of unit height and width 5 channels, centered in the bandpass, and cross-correlate this model spectrum with the real one. To do this, type,

```	>GET 1000
>HEIGHT=1; CENTER=H0(NOINT)/2+1; HWIDTH=5
>GMODEL COPY(0,1)
>#You have created a model of the line. Now cross-correlate
>#               model and spectrum and display results.
>GET 1000; CROSSFCN PAGE SHOW
```

Go to the previous, next section.

```Webmaster: Ronald J. Maddalena
```