If you are thinking about using FIR v6.1 (especially the MCV
architecture), then be warned. Seems like there's a bug with the FIR
6.1 core that is not documented in the errata sheet.
FIR settings:
- Multiple cycle variable (MCV) architecture
- 150 taps interpolating filter (contains both positive and negative
coefficients).
- 2 cycles/output
- No coefficient reload.
- Coefficients are stored in M4K
I inputted a delta function, expecting the FIR to output a delayed
version of the filter response. However, it seems like some outputs are
dropped while the others are repeated twice.
While there is an errata that seems somewhat related to the issue, but
the errata is only applicable when all coefficients are positive (and
MCV with no coefficient reloading). The suggested work arounds in
errata sheet are:
1. Use other architectures
2. Turn on coefficient reload, but disassert coef_we (to disable
reloading coefficient).
Since my FIR has two out of the three criteria, I thought I would give
the work-arounds a try... 1) works for me, but 2) still does not work.
Checked with Altera and this is indeed a bug. They promised to have
this fix in the next release that is coming out later this year.
Btw, the above error is also reproducable with other cycles/output.
