New Subjective Weighting Function ANSI C 63 19

New Subjective Weighting Function ANSI C 63. 19 Working Group Submitted for discussion by Stephen Julstrom January 19, 2008 1

Goal of the Subjective Weighting: • In the context of C 63. 19, the subjective noise-weighting function should predict the degree to which any measured level of undesired audio frequency pickup interferes with the desired use of the hearing aid. • Study results indicate that the degree of subjective interference is largely a matter of annoyance and not of intelligibility (i. e. , noise can be annoying even though the speech is still intelligible). • Analysis of the telecoil mode interference study data indicates that user annoyance and resultant acceptability correlate strongly with simple audibility of the interference (in the presence of speech). • From the results of the testing employing eight widely differing noise types, it can be said that A-weighting is a reasonable, but not ideal predictor of audibility, annoyance, and resultant acceptability. • In the following cumulative distribution graphs from the study report, the ideal weighting function would have given identical S/N results for all eight noise types (allowing for random experimental variation), with no consistent bias related to noise type. 2

Telecoil mode test results using A-weighting (In viewing the graphs, the 80 -90% range in the distribution is the most significant for comparison, since it includes the most subjects while excluding the outliers in the top few 3 percent. )

Indicators for a New Weighting Function: • A-weighting results in a roughly 10 d. B spread of S/N judgments among the noise types. An ideal weighting function would have the curves overlay each other (plus random experimental variation). • Comparing to the GSM curves, for example, UMTS and FHSS modulations appear to require higher S/N, while TDMA and Display noise appear to require lower S/N. • This is because, on a relative basis, A-weighting under-predicts the audibility of the former noise types and over-predicts the audibility of the latter. • An improved weighting function should increase somewhat the emphasis on the lower frequency content and the more impulsive nature of UMTS and FHSS, while de-emphasizing the high frequency noise character of TDMA and Display noise. (Interestingly i. DEN has some of both sets of characteristics, leaving it coincidentally well-represented by A-weighting. ) • In addition to the need to respond to the impulsive nature of some signals, a desire to measure the level of the interference while it is “on” (the WD transmission is often intermittent) further suggests the use of some form of quasi-peak detection (but without the extreme 6 k. Hz region emphasis introduced by ITU-R 468, for example). 4

Telecoil mode test results using the new weighting The previous ~10 d. B spread in the 80 -90% areas is reduced to just a few d. B, with much of that being random variation, rather than consistent bias related to noise type. 5

The weighting function: • The weighting function consists of straightforward spectral weighting followed by temporal weighting. • Each step is clearly mathematically definable. • Implementation may be in hardware or software. Spectral Weighting from squarelaw detector RMS Level Measurement X 2 to DC meter X 0. 5 4 msec TC Similar to A-weighting Peak Detector 550 msec T. C. Similar to CISPR, ITU-R 468 Quasi-Peak Detection 6

The spectral weighting portion: from squarelaw detector (or conditioned baseband magnetic probe for ABM 2) x 2 20. 6 Hz 12. 2 k. Hz 737. 9 Hz x 1. 2589 (A-weighted) 107. 7 Hz 369 Hz 3 k. Hz, 2 nd order D. F. = 0. 707 x 1. 0864 new-weighted 7

Justifications for the specific deviations from A-weighting: • Slightly enhanced lower frequency sensitivity: Justifiable on the basis that the interference will typically be heard at a 40 to 50 phon level, rather than the 30 phon level that was the original basis for A-weighting. • Significantly reduced high frequency sensitivity: Justifiable in light of the probable uncorrected or uncorrectable high frequency hearing loss of the test subjects at the typical noise levels in the presence of typically 65 d. B-SPL equivalent speech. • Temporal weighting (a form of quasi-peak detection): Justifiable in light of the ear’s probable sensitivity to impulsive sounds with low average energy. While the quasi-peak detection defined here is not dissimilar to that found in related CISPR and ITU-R 468 standards, it is more simply and clearly mathematically defined, works well in combination with the spectral weighting for the test signals evaluated, and does not need to be applied directly to the RF waveform. • The major justification: The weighting correlates consistently to test subjects’ subjective judgments over a wide range of noise types. 8

Applicability to RF Interference: While the data leading to the new weighting function came from the telecoil mode study, the weighting should have equal applicability to RF interference. Seven of the eight baseband audio frequency magnetic noise types studied had their origin in the envelope of the RF amplitude modulation, resulting in a very similar audible character to related RF interferences: • The square law detection of RF interference vs. the linear pickup of baseband magnetic interference does not result in spectra of such a different nature that they would likely contradict either the study results or the alternate weighting derived from them. In the case of pure pulse waveforms such as GSM, FHSS, DSSS, and CDMA (within the audio band), the recovered spectrum is not changed at all by square law vs. linear detection. • The direct magnetic interference pickup effectively applies a “telecoil response” to the RF modulation spectrum; that is, a 1 st order rolloff below 1 k. Hz. Thus, RF interference will have relatively somewhat more lower frequency content, in comparison to the corresponding signals as heard by the subjects in the telecoil mode study. The difference is fairly small, though, within the limited lower frequency range of potential audible interest, and should be adequately accounted for by the frequency weighting. Again, the difference is not so great as to likely contradict the study results or the derived weighting. 9

Addendum: New weighting vs. A-weighting comparison: The table below shows the shift in d. B resulting from measurement of the various noise types using the new weighting in comparison to A-weighted measurements. (Aweighting is not presently employed in the RF measurements. ) The results were modeled using the continuous noise recordings of the telecoil mode study. • The “T-coil response” column numbers were subtracted from the corresponding curves in the original study data of slide 3 to obtain the graphs of slide 5, giving the test results as they would have occurred had the noise samples been normalized using the new weighting instead of A-weighting. • The d. B shifts of the “RFI response” column can similarly predict the difference of the two weightings, assuming testing otherwise according to the proposed generalized method. 10