The most natural thing to do when considering this kind of generic maturity analysis of bonds is to use a similar series of bonds to derive a bond curve which represents a single yield curve that reprices them minimising the least squares error (you want a reasonably smooth curve that doesn't necessarily price them all precisely but captures the generalist structure).
This way you build a system whereby you can derive any price you want for any maturity at any time (within the error imposed by your system).
Let's say you went back to 1998 and priced a bond with maturity of 2018, in that instance the bond would represent a 20Y and your constructed curve in that sector would reflect prices of the 20Y bonds at that time.
Note that you can't directly combine time series of different bonds because each bond has inherent characteristics that make it slightly cheaper or slightly more expensive and therefore when you merged the timeseries you have to account for that difference, which would no doubt introduce more error that the method above and be far more subjective and arbitrary for research purposes.