The “Peak-Picker”

If Necessary, Can We Reduce the Amount of Information Communicated, and Reduce Inter-Electrode-Channel Interactions, by Choosing a Subset of Channels to Stimulate during each Time-Frame?

Quotes from the Literature:

• The original concept of the peak-picker came from Peterson E. and Cooper F.S., 1957, "Peak-picker: A Bandwidth Compression Device," JASA 29:777 (A).  The second author should be of particular interest to the cochlear implant community because Frank Cooper made contributions to cochlear implant research, in addition to his well-known contributions to speech science.   He was a founder of Haskins Laboratory at Yale.  Our laboratory at UCSF had the benefit of Frank's advice an support a number of times: My first experience with Frank was when he was the chair of our first NIH site visit -- for the first NIH-funded research for testing cochlear implants in humans.  He was very knowledgeable, interested, and kind.  It was inspiring to realize that Frank ended-up helping to develope major "regions" of speech science as a result of his original (and continued) interest in developing a reading machine for the blind!  "An engineer turned scientist" for two wonderful reasons: (1) to benefit his fellow man and (2) to try to understand a part of nature.

• The peak-picker strategy is better viewed as a "processor-modification-strategy" or a "processor-add-on-strategy" since it can be applied or "added-on" to most multi-channel cochlear implant processing strategies.

• Description of the peak-picker method and a suggested use in cochlear implants: excerpt from page 80 and the excerpt from page 81-83 of White, 1978.

• Use of peak-picking (and a "soft-form" of peak-picking) to reduce channel interactions in cochlear implants: excerpt from page 501 of White, Merzenich, and Gardi, 1984.

• The previous excerpts presented two "uses" for peak-picking:
  
  (1) Peak-picking to reduce the "information-load" on the CNS -- by only stimulating a subset of channels, presumably those channels that convey the most information.  For this rationale to of value, some patients' CNS may receive speech better when only "attending to" the most important information.  This may also be viewed as a form of feature detection/extraction, or "information-importance" detection/extraction.  In that sense, this "use-case" involves feature extraction. -- ["use-case" is software-geek-lingo -- sorry but I like the term!]
  
  (2) Peak-picking to "directly" reduce inter-channel interactions -- by reducing the number of driven electrode channels; or by reducing the stimulus amplitude driving some channels.  Because fewer electrodes are receiving higher amplitude currents, the likelihood that the electric fields will interact (at the nerve) is reduced. The number of channels that "are picked-from" can be very high thereby allowing potentially very high frequency resolution. -- however the number of channels, that are driven heavily, is limited.  So, the effective spatial resolution of the system could be quite high -- depending on a number of factors!
  

• Interestingly, the peak-picker strategy was first applied to an Analog speech processor (see: excerpt from page 11 of the 7th QPR from RTI (1985)).

• The first implementation of the CIS processor did not include the peak-picker modification/add-on (see the implementation section for information about the UCSF CIS processor).

• A peak-picker strategy was applied to an IP speech processor, which was implemented by RTI and tested at UCSF (see: excerpt from page 13 of the 7th QPR from RTI (1985).

• RTI renamed the peak-picker strategy "the n-of-m processor:"  see excerpt from page 3 of QPR 1 (1995).  Also, see QPR 2 (1995) and QPR 3 (1996).  RTI originally called the "m of n" the "peak-picker."  Why did RTI change the name, particularly since the original name, "peak-picker," was historically accurate both in origins and concept. Furthermore, "peak-picker" communicated the algorithm's operation better IMO.

• 1999 Letter to the Hochmairs by M White: Letter with referenced excerpts, excerpt 1, excerpt 2, and excerpt 3 from White, 1978; and the excerpt from Bruce, et al, 1999.

• In 2000, RTI describes the "n of m" processor in interesting detail: see the excerpt from page 37 and the excerpt from page 38 of the 5th RTI QPR (submitted to NIH in 2000).  IMO, some of the material in these excerpts appears to be some sort of circular rationale, and perhaps even some slight concession, related to their renaming of the peak-picker algorithm.

• One of the reasons that many investigators are unaware that RTI renamed the peak-picker strategy is due to the availability of RTI's quarterly progress reports:  None of RTI's QPRs for Contract # N01-NS-9-2401, covering the period 1988-1992, were available on NIH's web site -- until recently: click-here to see a "snap-shot" of the previous web site's contents.  All other RTI QPRs were available (QPRs before and after this 1988-1992 period) through the web site.  After I made a request, a new staff member at NIH made all of RTI's QPRs available. However, recently, they were unexpectedly removed from the NIH website. Fortunately, in 2004 I had copied the reports from the NIH web site and I have made the QPRs available at this website.