I wrote a program called LocusPocus that uses gene annotations to break down a genome into gene loci and intergenic regions. It handles overlapping annotations and other weirdness pretty robustly. The output will include both coding regions and non-coding regions, but you can identify the intergenic spaces as those with iLocus_type equal to iiLocus or fiLocus.

Note: the --delta parameter will extend each gene/transcript by 500bp by default.

Caveat: the program only accepts GFF3 input by default. Hopefully it won't be too hard to convert your GTF to GFF3.

This isn't a problem that's easily solved with awk. It's not like you're extracting a feature that's annotated in the GTF file. Instead, you want the empty space between annotated features.

A few years ago I wrote a program called LocusPocus for a similar task. It uses a gene annotation to break down a genome into gene loci and intergenic regions. It handles overlapping annotations and other weirdness pretty robustly. The output will include both coding regions and non-coding regions, but you can identify the intergenic spaces as those with iLocus_type equal to iiLocus or fiLocus.

Note: the --delta parameter will extend each gene/transcript by 500bp by default.

Caveat: the program only accepts GFF3 input by default. Hopefully it won't be too hard to convert your GTF to GFF3.

Another caveat: eventual interpretation of these data will depend on what features are annotated in the genome and which annotations you include vs ignore. Do you want your non-coding regions to include non-coding genes, or should these be treated separately? Some non-coding regions will be full of transposable elements and other repetitive DNA, while others will have enhancers, promoters, or other regulatory elements. It's important to tread carefully before you jump to any conclusions.