In this example a digital audio stream is upsampled by ratio 3 in order to transform a recording from 32kHz consumer DAT to 96kHz in order to prepare it for post processiong. This article shows some technical background, why this is usefull: Do we need 96kHz ?

Insertion of Zeros

From the theory it is sufficient to insert zero values for non defined samples when creating the higher data rate and apply a filter. Pratically this has some issues:

Depending on the subsequent reconstruction filter, a non ideal reconstruction is achieved especially with BIAS loaded data streams. This is because of the non ideal filter curve which never meet the requirement of a perfect cut off frequency.

In this example a linear approximation in between the given data points is used to defined the missing values.

Here a dynamic interpolation filter is used to reconstruct the non existent values. It is a modified newton filter. In this case the data stream has less requirements regarding the reconstruction filter.

Conclusion and Summary
Intelligent interpolation filters can be used to support digital upsampling creating less artefacts.