That wouldn't be a problem. Note that I did write 4-8kB of RAM besides the measurements data. The figure was for bitmap memory.

With 1k horisontal resolution (8 inches wide paper and 120 DPI), and a 24-pin printer head, a single head pass would require 3k of RAM.

In this case, the printer has a 9-pin head. 48 bit * 120 DPI * 8" would require 5760 byte and allow for 5 head passes.

There really is no need to consider the total amount of data to store a full image. A dot-matrix printer only cares about row-by-row output. And even switching to a page printer makes no much difference - a laser printer would have enough memory to wait until the full graph has been generated and transmitted. An ink-jet printer may be problematic, because delays between multiple head passes may affect the print quality.

I did a lot of similar work once on a Z80 machine with 16kB RAM. The Z80 is a bit more general-purpose but a 8051 shouldn't have a problem either, as long as the OP has a very basic knowledge of computer graphics. No problem at all to produce full-page graphs.

But even when moving to the PC, early MS-DOS and Win 3.1 applications normally had to do banded output since the 640kB of RAM was to a large part crammed by the application(s), leaving a minimum of RAM for data.

Settling for a bar graph would make the task trivial, since no computer-graphics knowledge would be needed at all. But almost any form of graph can be generated with very little RAM and a minimum of compuations - most draw primitives would use integer DDA - Digital Differential Analyzer - i.e. all you need to do is to compute integer differences, make a comparison and decide to move horisontally or vertically for your next pixel. So you can to a large part avoid multiplications, divisions, or trigonometry. The algorithms can be googled for, so you don't even need to understand the math behind their operation. And any entry-level computer graphics book will contain the algorithms together with an explanation how they work.

In the end, I don't think you would bee too interested in seeing the code. This is just one more example of a problem that may sound hard until you know the solution. Then it suddenly becomes borderline trivial. The graphing primitives are quite small, but you would add a clipping layer where the code runs through all measurements data and only draws the part that is visible in a single band. After ejecting the band, you draw and emit the next band until the full graph has been printed. Then you finalize the output with a form feed.

Some clipping is best to do before/above the drawing primitives. Some clipping may be better done below the primitives. If you draw a perfect circle using DDA, you may decide to clip on-the-fly when drawing. If you draw a circle as a number of straight lines, then you can perform the clipping before drawing the line segments.