1、IBOC ANALOG-DIGITAL BLEND TIME ALIGNMENT QUALITYMichael BergmanKenwood CorporationJohn KeanNPR Labs - National Public RadioABSTRACTAnalog-digital blend is a unique feature of HD Radio. Proper time alignment of the audio chains appears to be a key factor for good performance in signal fading conditio
2、ns. Poor alignment leads to audible level shifts during blend, echo, and comb-filter artifacts. This paper looks in depth at time alignment measurement techniques and the resulting accuracy, presents IBOC receiver alignment results across several brands, and considers audio quality issuesBACKGROUNDI
3、n the United States, and increasingly in other parts of the world, terrestrial digital audio broadcast is done using NRSC-5-A IBOC Digital Radio Broadcasting Standard (the National Radio Systems Committee specification for In-Band-On-Channel transmission). Currently, the only implementation of this
4、is HD Radio, which comes from iBiquity Digital Corporation, the technologys creator. IBOC carries digital audio via OFDM modulation, but also allows for simultaneous transmission of legacy analog AM and FM. NRSC-5-A IBOC has the unusual ability to “blend” between analog host audio and Primary Digita
5、l audio as reception conditions warrant. For the FM band, in multipath fading environments, the receiver smoothly switches from the OFDM-transported digital audio to legacy FM analog audio until conditions improve; it then blends back to digital. This blending is accomplished by synchronizing the an
6、alog and digital audio waveforms at the transmitter, carefully calibrating the transmitters analog-to-digital time alignment, taking into account a receivers standard analog-to-digital time alignmentall to result in a maximum time differential between the analog and digital of 68S.This explanation b
7、egs several questions: What can be used as a standard reference for calibration? How does one calibrate hardware at each end? What calibration accuracy is required at each end? How does calibration vary with time or manufacturing process? What happens if calibration is off? Through a series of inves
8、tigations at NPR Labs, quite a bit of detail on these questions comes to light.CREATING A REFERENCECalibration for anything presupposes the existence of a calibration standard. When iBiquity established its standard for time alignment it was realized that it could not be made traceable back to NIST
9、as it is a relative, not absolute, standard. Time alignment depends on matching the delay through the analog path to the delay through the digital path. These paths start with the audio feed into the exciter or audio processor, and end with the receiver audio output. Since both paths have to go thro
10、ugh the IBOC technology, there isnt a single fixed reference. The analog path includes exciter processing and a 7 second delay; the digital path includes a great deal of buffering and processing. The only way to have a standard reference is to create one and then ensure that each transmitter and rec
11、eiver works within specification.Originally, iBiquity created a test vector. This vector is a coded fixed signal that can be played back by a signal generator. The over-the-air RF signal is therefore very consistent and can be used as a fixed reference for receiver designers. They used this to calib
12、rate the iBiquity reference receiver. Next, iBiquity and Kenwood calibrated the original Kenwood KTC-HR100. With a calibrated receiver reference, the transmitters analog delay could then be determined. To ease this, many receivers have a special mode which splits the stereo outputs, with the stereo
13、left channel output typically carrying digital audio left channel signal, and the stereo right channel output carrying analog audio right channel signal. Time delay between the digital left and analog right waveforms can then be compared.THINGS GOING WRONGThis “split-mode” test assumes that the orig
14、inal audio was suitable for a tight calibration. As a trivial example, if the digital content does not match the analog content at all, this method will not work. For example, if the analog audio path carries a hip-hop track, and the digital audio track carries jazz, there will be no way to match wa
15、veforms. (This example is of course academic, since the FCCs authorization for IBOC requires that broadcasters simulcast the main channel audio on the analog and digital signals.)A more likely example occurs when the audio processing varies between the two paths. Experience with measurement of actua
16、l stations indicates this is a common issue. Waveform phase can vary severely if the analog and the digital paths go through the separate processors.Assuming simulcast audio, and joint audio processing (analog and digital are jointly processed in a way that ensures some level of phase matching), we
17、can consider calibration of the equipment.TRANSMITTER CALIBRATIONThere are several methods available, each with its own characteristics and with varying accuracy:1. Calibration by Ear2. Calibration by Test Vector3. Calibration by WaveformThe easiest method is calibration by ear. The broadcast engine
18、er simply listens to the audio, forcing blends back and forth. This is not recommended. The second method, by Test Vector, requires test data from iBiquity and a signal generator. Since this is not commonly available broadcast equipment, the third method, Calibration by Waveform, is preferred.The re
19、st of this paper investigates the method, challenges and accuracy of calibrating using waveform analysis.TEST METHODOLOGYFirst we calibrated a test Harris Dexstar exciter with a Kenwood KTC-HR100TR IBOC receiver. The method for this was to assume the receiver was already calibrated, which the manufa
20、cturer indicates is the case. The procedure included the following steps: Put test receiver in split modeWe put the Kenwood receiver into “split mode” (analog out of the right channel, digital out of the left). Split mode on the Kenwood product is enabled with a key sequence available from the manuf
21、acturer or in an application note at . Use a L+R tone burst as audio source materialA sinusoidal rectangular burst tone (1000 Hz tone burst, 5 ms duration, 5 sec interval) is a good alignment test signal for HD Radio and analog host FM channels. This audio file is Track 99 of the “NAB Broadcast and
22、Audio System Test CD”, available from the NAB Store online. Capture audio waveforms from the receiver using a high-quality audio input samplerA Lynx professional PC sound card was used to digitize the analog audio from the receivers under testLess-costly PC sound boards (internal, standard line-in h
23、ardware) may be usable but must be validated by the user for stability. For example, one laptops line-in hardware exhibited poor audio SNR in the recordings. One laptops line-in hardware was mono, not stereo. Some inexpensive audio input hardware samples right and left channels alternately, not simu
24、ltaneouslyWe therefore concluded that there may be some laptop or desktop PCs with clean stereo line-in hardware, but engineers should not assume that their PC qualifies as such. A CD source with the 1kHz tone burst fed directly into the PC should be used to verify that the PC audio system is high e
25、nough quality. View the captured audio in a sample-level waveform editorIn our setup, we used Cool Edit 2000 to view waveforms, adjust audio levels and measure timing differencesThis overall approach is a good indicator of time alignment, providing resolution down to one sample (1/44,100th of a seco
26、nd or approximately 22.7 S). Make processing as close as possible to real-life, but watch for processing-induced issues with testing.We used an Audemat-Aztec FMX410 stereo generator to provide the appropriate 75 S pre-emphasis and group delay vs. frequency in the analog FM channel. We found that the
27、 generators audio processing caused a slight amount of clipping on the tone bursts. The audio processing was turned off to eliminate the clipping.With the transmitter at a known point relative to a (presumably) calibrated receiver, we were able to begin testing more receivers.We choose the initial r
28、ise time point of the burst waveform as the place to measure. iBiquity uses this in their specification for such testing; however, they specify a digital oscilloscope as mentioned above. Figure 1 shows the digital (upper) and analog (lower) channels of a calibrated transmitter-receiver pair. The ver
29、tical black line shows the initial rise time point of the waveforms. This is the basis of the “rise-time” method of blend alignment. RECEIVER-TO-RECEIVER VARIATIONSWe looked for receiver-to-receiver variation in analog-digital blend time alignment within a single model (manufacturing process variati
30、ons in alignment). That is, does a single receiver model calibration hold for all the units coming off the production line? We tested four Kenwood KTC-HR100TR units, built over a 10-month period, to see if they had similar calibrations relative to our calibrated transmitter. Table 1 shows the result
31、s when the calibration of a single model receiver is checked over multiple units at a given time. This tight match was repeated three times during testing. In this table, “Samples analog lags digital” refers to how many 44.1 kHz samples (about 22.7S) that the analog audio is behind the digital. The
32、iBiquity specification for alignment is 68 S; this is about three audio sample times.Model S/N Samples analog lags digitalKenwood KTC-HR100MC/TR# 30900166 0Kenwood KTC-HR100MC/TR# 30900867 1Kenwood KTC-HR100MC/TR# 40700143 0Kenwood KTC-HR100MC/TR# 40700157 0Table 1: Receiver-to-Receiver variation te
33、st resultsIn this test, we are not comparing receivers tested hours apart due to concerns about transmitter jitter. For a particular point in time, receivers from a common design remain within one 44.1 kHz sample of each other. Of course, this is not exhaustive testing. However, iBiquity maintains t
34、hat the receivers should in theory be identical across production lots, so this was simply confirmation that receivers do not vary across production lots.MANUFACTURER-TO-MANUFACTURER VARIATIONSWe next looked for manufacturer-to-manufacturer variations in receiver blend time alignment. That is, are a
35、ll manufacturers calibrating to the same standard? In addition to the Kenwood model, we tested models from JVC, Boston Acoustics, and a professional receiver from Day-Sequerra. Table 2 shows the summary results for five receivers and a monitor. Each of the receivers was well within the iBiquity spec
36、ification of 68 S (three samples). Model S/N Samples analog lags digitalKenwood KTC-HR100MC/TR# 30900166 0JVC KD-HDR1 # 119X0336 0JVC KD-HDR1 # 139X0172 0Boston Acoustics Recepter# AFQ 5D001544 -1Boston Acoustics Recepter# AFQ 5D002623 0Day-Sequerra M4 # D70002 0Table 2: Receiver-to-Receiver variati
37、on test resultsWe concluded that for the receivers tested there is no significant manufacturer-to-manufacturer variation. As Figure 1: Digital (upper) and analog (lower) output from a calibrated transmitter-receiver pair.a result, any of these receivers could be used for calibration.TRANSMITTER TIMI
38、NG JITTERTable 2 is a summary of the “typical” results. What is not shown in Table 2 is that occasionallyalbeit rarelywe did see up to 2 samples, rather than the worst-case value of -1 in the table. That is, we would sometimes see a value of +2 or -2 in our testing.At certain times during the day, a
39、ll receivers would shift one or two samples to the right or left. This added variation appears to be due to the exciter. That is, the receivers remained quite consistent to each other, but all would simultaneously shift one or two samples. As a result, the data varied slightly during the day, even f
40、or a single receiver tested in the morning compared to the same receiver in the afternoon. This was observed on two separate days.Since all receivers of that design behaved the same way at any particular time, we concluded that most likely the transmission chain has one or two samples worth of delay
41、 variation, or large-scale timing jitter, exhibited over a span of hours.RECEIVER ABSOLUTE LATENCYThe tested receivers had consistent RF-in-to-audio-out latency. That is, assuming the delay through the transmitter was consistent, the delay through a receiver was always the same. However, different r
42、eceivers were not equal in this respect. In fact Kenwood and the Boston Acoustics (BA) Recepter Radio HD differed by 5120 samples (116 mS). This measurement was very repeatable; three successive tests were run.The first (baseline) test involved measuring the RF-audio latency between the Kenwood (dig
43、ital) audio out, left channel; and the BA (digital) audio out, right channel.The second test was identical to the first, except that the BA was powered down and back up before running the second test.The third test was identical to the second, except that the Kenwood was power-cycled first.In all ca
44、ses, the difference between the Kenwood and BA receivers was the same: exactly 5120 samples.We did not find any variations in receiver latency from boot to boot, based on comparing the Kenwood to the BA. However, absolute latency through a single receiver is not guaranteed in generalonly the relativ
45、e latency from analog to digital (i.e., blend time alignment) is guaranteed.This means that it is inadvisable to compare a receiver in digital mode to an identical model in analog mode for blend time alignment testing. Instead, the test should be performed with one receiver in split mode. MODULATION
46、 MONITORS AND REAL-TIME ALIGNMENT IN THE LABIdeally, the broadcaster would like to be able to monitor blend time alignment in real-time, off the air. Several monitor makers have incorporated correlation-based time alignment monitors into their products. We tested the Audemat-Aztec Golden Eagle FM-HD
47、 and the Belar FM-HD1 modulation monitors.Our first tests were to verify the monitors under lab conditions. We began with minimal processing enabled, a stereo music source, and a calibrated Dexstar. We observed the effect of changes in blend time alignment on the Belar FM-HD1.It appears that even sm
48、all adjustments in timing offset (performed by adjusting the Dexstar) such as 0.16 mS (about 7 samples) were sensed by the Belar. Moreover, it took the Belar only a few seconds to resolve these small errors. The Audemat performed almost identically in all cases.Frankly, this was not a real-world cha
49、llenge. The program material was an unprocessed cello solo performance, which provided identical audio for both HD and analog. Processing was minimal. There were no channel artifacts, since this was a lab test with no impairments. The next test was to introduce real-world conditions.OFF-AIR ALIGNMENT METHODOLOGYPreviously, we described a test methodology for lab alignment, using a tone burst mono signal. To consider off-air alignment, we needed to
