Reply by Rune Allnor July 22, 20082008-07-22
On 22 Jul, 00:35, Manny <mlou...@hotmail.com> wrote:
> Hi, > > I'm working on an application for TOA and TDOA for a set of sensor > data. After matched filtering, I'm running my data through a 39-tap > Hilbert Transform to ultimately obtain the signals' envelope and store > the in-phase and quadrature components of the peaks for subsequent > processing. > > Due to the filter latency, very close sources are not detectable. So I > thought of using circular buffers whose lengths are equal to the > Hilbert FIR filter and stitch these to my real data set; effectively, > providing what I think is a reasonable workaround for the latency > problem. > > I was wondering whether this approach makes sense to the DSP veterans > here and whether there exist out there a neater solution.
The resolution problems have nothing to do with filter latency, regardless of whether the sources are close in angular domain or time domain. Whatever resolution issues you struggle with, first have a look at the time-bandwidth product and the Heissenberg inequality. These are fundamental properties which link resolution to the framelengths (or array lengths) of data under analysis. Rune
Reply by Manny July 21, 20082008-07-21
Hi,

I'm working on an application for TOA and TDOA for a set of sensor
data. After matched filtering, I'm running my data through a 39-tap
Hilbert Transform to ultimately obtain the signals' envelope and store
the in-phase and quadrature components of the peaks for subsequent
processing.

Due to the filter latency, very close sources are not detectable. So I
thought of using circular buffers whose lengths are equal to the
Hilbert FIR filter and stitch these to my real data set; effectively,
providing what I think is a reasonable workaround for the latency
problem.

I was wondering whether this approach makes sense to the DSP veterans
here and whether there exist out there a neater solution.

Would appreciate your input on this.

Many thanks,
-Manny