[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

*To*: AUDITORY@xxxxxxxxxxxxxxx*Subject*: Re: human versus spectral resolution*From*: Yi-Wen Liu <jacobliu@xxxxxxxxx>*Date*: Thu, 3 Apr 2008 10:08:48 -0500*Comments*: To: Michael Fulton <michaeljfulton@xxxxxxxxxxx>*Delivery-date*: Thu Apr 3 11:12:10 2008*In-reply-to*: <BAY103-W24A2D2BF162B9AA4EBD372C8F40@xxxxxxx>*List-archive*: <http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>*List-help*: <http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>, <mailto:LISTSERV@LISTS.MCGILL.CA?body=INFO AUDITORY>*List-owner*: <mailto:AUDITORY-request@LISTS.MCGILL.CA>*List-subscribe*: <mailto:AUDITORY-subscribe-request@LISTS.MCGILL.CA>*List-unsubscribe*: <mailto:AUDITORY-unsubscribe-request@LISTS.MCGILL.CA>*References*: <BAY103-W24A2D2BF162B9AA4EBD372C8F40@xxxxxxx>*Reply-to*: Yi-Wen Liu <jacobliu@xxxxxxxxx>*Sender*: AUDITORY - Research in Auditory Perception <AUDITORY@xxxxxxxxxxxxxxx>

Hi Machael and all, > > 1) Time/frequency resolution for digital signal spectral analysis , as (sort > of) governed by Heisenberg's uncertainty principle with relation to the > duration, sampling rate, frequency content and bandwidth of the sound. > > 2) The limits of the ability of the human ear to distinquish between > frequency/pitch and the exact time location of sounds, more or less the same > task as above. I'd like to point out there are two notions of uncertainty regarding frequency estimation. One is the uncertainty due to the presence of multiple sinusoids and limited observation time, and the other is the uncertainly due to the presence of noise. The first kind of uncertainly is governed by Heisenberg's principle, the second kind can be assessed by the theory of Fisher information and Cramer-Rao inequality. Heisenberg's principle here can be written as (Delta f)(Delta t) >= 1 (maybe there is a factor of 2pi?) which says that the best frequency resolution is in the order of 1/(Delta t). In other words, two sinusoids of frequencies f1 and f2 are NOT well-resolved within Delta t < 1/(f1-f2). Fisher information considers the limit of parameter estimation under noisy observation. When applied to frequency estimation under additive white Gaussian noise, the theory predicts that the variance of frequency estimation error (Delta f)^2 will always be greater than the Cramer-Rao lower bound, which is proportional to SNR/(Delta t)^3. For a thorough discussion of Cramer-Rao bound in sinusoidal parameter estimation and some estimators to achieve the bound, check Boaz Porat (1993). Digital Processing of Random Signals, chapter 9: "Estimation of deterministic processes". For a compact and elegant discussion of Fisher information, you can check Cover and Thomas (1991) Elements of information theory, Chapter 12 "Information theory and statistics" An early paper on this topic is Single tone parameter estimation from discrete-time observations - D Rife, R Boorstyn - Information Theory, IEEE Transactions on, 1974 - Vol. 20, Page 591-98. Best regards, Yi-Wen

**References**:**human versus spectral resolution***From:*Michael Fulton

- Prev by Date:
**Re: human versus spectral resolution** - Next by Date:
**Identification of musical tones** - Previous by thread:
**Re: human versus spectral resolution** - Next by thread:
**Re: human versus spectral resolution** - Index(es):