DV.com

DV.com DV.com - Inspiring and Empowering Creativity

Jay Rose and Dan Rose

DV Camera Audio: Real Numbers, Real Recommendations

We all know the DV revolution changed the face of filmmaking. Anybody with skill, a few thousand dollars, and a computer can make a film that looks good enough for network broadcast. However, the audio quality of these cameras doesn't equal the video, and economics suggests this probably won't change. These cameras are aimed at high-volume sales, so they have to be built on a tight budget. A few extra dollars of cost per unit can make a serious difference in the manufacturer's profit. Because people pay the most attention to picture when buying a camera, sound quality is inevitably compromised. This isn't just grousing by a couple of audio curmudgeons. To take only one example, go to Sony's Web site (http://bpgprod.sel.sony.com), and see what Sony considers important. For its audio equipment, Sony brags about frequency response, harmonic distortion, and noise-things that indicate how well a circuit will sound. But when you look for audio specs on a camera such as the PD150, all you find is "48 kHz/16 bit, 32 kHz/ 12 bit." This tells you as much about sound quality as NTSC video tells you about the CCDs or lenses. Even Sony's marketing department couldn't supply us with deeper technical details.So we had to gather our own infor-mation. We took some popular DV cameras and a few audio alternatives, and hooked them up to sophisticated test equipment. We didn't look at every camera available, but we chose a good sampling of cameras at different price points. This article explains what we discovered. The first part covers how we tested, and the second part reveals the specific numbers for individual cameras (many of them were disappointing). The third part suggests ways to get the best sound possible, despite the numbers.

Dan Rose gets ready to start checking a camera. We connected the cameras' FireWire output through a Mac G4 with AES/EBU, then sent the output through a studio-quality digital-to-analog converter to an Amber 3501 distortion test set and Terrasonde ATB+ analyzer.

Behind our tests

To measure camera audio performance, we looked at three areas: noise, distortion, and frequency response. All of our tests fed calibrated signals to the cameras' analog inputs. Their outputs were connected via FireWire through a computer with AES/EBU outputs, to a studio-quality digital-to-analog converter. They were then measured in a well-maintained Amber 3501 distortion test set and a new Terrasonde ATB+ multifunction audio analyzer (www.terrasonde.com). We also used an oscilloscope to examine distortion products. Bypassing the camera, our setup had 0.01 percent THD+N (Total Harmonic Distortion + Noise), better than 80 dB s/n (signal-to-noise ratio) below -12 dBFS, and was within 0.25 dB from 20 Hz to 20 kHz.Our tests involved mic-level signals at -35 dBu-about what you'd expect from a good mic close to the subject-using the MIC ATT position when available. We also tested at -50 dBu, representing more marginal miking, and at line level (-10 dBV for most devices; +4 dBu for the DSR-570). We used 48 kHz sampling for every test except those on the MiniDisc, which is preset to 44.1 kHz.

This chart details the measurement of each camera's and recorder's Total Harmonic Distortion + Noise (THD + N), a measure of sound purity. Shorter bars are better. The yellow bars indicate reasonable quality when used with a medium output mic or an external preamp. The three tall blue bars represent cases where the manufacturer probably traded distortion for noise at very low levels.

Noise

The s/n is a measurement, in decibels, of how much hiss and other electronic junk a circuit adds to a recording. A higher s/n means less noise and a cleaner track.Signal-to-noise can't be measured without a reference level for the signal. In analog, this is usually 0 VU, although momentary peaks frequently exceed that level without causing problems. But in digital, 0 is an absolute limit. Signals that exceed it get horribly distorted. Most miniDV camera manufacturers recommend an average level of -12 dBFS for dialog, with the rest reserved for a safety margin as headroom. So we used that level for our digital measurements, which also means the numbers can be compared with traditional analog equipment.Although 16-bit sound is capable of 84 dB s/n below -12 dBFS, that kind of quiet is almost never achieved in practical equipment. If you're shooting for theatrical release or heavily processed broadcast, 60 dB is desirable. Corporate and event videographers can get by with 45 dB or so if the voices don't have much dynamic range; more is needed if there are volume swings.

The signal-to-noise ratio (s/n) is a measurement of hiss and other junk. Longer bars are better. Despite various rumors, the PD150 tested almost the same as the original XL1. We expect the newer XL1S's performance will be closer to that of the GL2.

Distortion

When an audio circuit's output has tones that weren't in the input, that's distortion. Back in the analog days, distortion usually took the form of extra harmonics at musically related frequencies. Digital distortion can be uglier and nonmusical. A distortion analyzer looks for and measures anything that's not part of the input-analog-style harmonics, digital aliasing, random noise. The measurement is called THD+N (Total Harmonic Distortion + Noise). It's expressed as a percentage of the total output.The THD+N measurement can reveal noise that's been otherwise hidden by noise gates. In some equipment, a digital circuit turns off the output when the input goes below a preset value. This turns off any circuit noise (as well as any signal), making the device appear quieter in a traditional s/n test. But because the gate is open during a THD+N measurement, the noise shows up as increased distortion. Distortion and noise are also related in a circuit design, and manufacturers can trade them off to a limited extent. So if you see a camera with particularly good ratings in one regard, make sure the other hasn't been compromised.A THD+N of 1 percent used to be acceptable in pro analog recording. These days, 0.1 percent is considered reasonable, and a lot of pro gear is better than 0.01 percent. But 1 percent is still adequate for dialog, particularly in a FireWire-connected editing situation where subsequent transfers won't add distortion.

The position of a device's volume control affects the s/n measurement. Longer bars are better. Read the individual camera measurements for some implications.

Decibels

Decibels don't have intrinsic values like volts or watts. They're logarithms, mathematical shortcuts for expressing ratios. A ratio of 1:1 is 0 dB. A ratio greater than 1 yields positive decibels, so 2:1 equals 6 dB. Negative numbers mean the ratio is less than 1 (1Ú2, or 1:2, equals -6 dB). Decibels can be used by themselves to compare two voltages, such as the ratio of signal to noise or how the level changes between 20 Hz and 1 kHz. But they're often used to compare a single signal with a standard reference, identified by a letter or two after the dB abbreviation. Condenser microphone output is measured in dBu, its ratio to 0.775 volts. Digital signal levels are measured in dBFS, a ratio to signal at full scale, or how loud things would be if every bit in the digital word was turned on.THE RESULTSWe tested each camera and recording device in a similar manner. S/n ratios are relative to -12 dBFS, the recommended operating level of most prosumer cameras. We turned MIC ATT (mic attenuation) on for the -35 dBu signals-it really does improve the sound-but we had to use the normal mic position to get a proper level from the -50 dBu signals. To level the playing field, we used balanced inputs for the mic signals. Volume controls were kept near their midpoints for most measurements. Now let's talk about the specific results of the cameras and recorders

Here we see frequency response for two typical prosumer cameras, a professional camera, and a portable DAT recorder. The devices have adequate response for dialog recording, but the DAT recorder is obviously better for music. Frequency specs for every device tested are in the main text.

Canon XL1

We tested the Canon XL1 (www.canondv.com) with a variety of inputs: the built-in RCA connector, Canon's $250 MA100 XLR adapter, and Beachtek's $170 DXA-4c XLR adapter (www.beachtek.com). We didn't test the newer XL1S, but we did look at Canon's new GL2 (see the following section). On the XL1, frequency response was down 2 dB at 80 Hz and 12 kHz through the electronically balanced MA100, and down 2 dB at 90 Hz and 10 kHz through the transformer-equipped DXA-4. The MA100 was down 5 dB at 15 kHz. The headphone output was down only 2 dB at 15 kHz, suggesting it is connected before the analog-to-digital converter.We found moderate aliasing distortion, a digital gremlin that adds nonmusical whistles and whines to harmonics of high-frequency signals that approach half the sample rate. Pro audio equipment reduces this problem by using oversampling. The XL1 added multiple tones between -25 and -30 dB for signals in the 500 Hz to 10 kHz range, and single tones -20 dB for signals 12 kHz and up. This told us the camera doesn't use oversampling, which also suggests the high-frequency loss is due to the steep filters the camera needs to reduce aliasing. It also suggests there's more serious harmonic distortion at higher frequencies. But the XL1 wasn't alone in this area.The position of the volume knob can have a strong effect on noise. So does the balancing adapter's design: Electronic ones have better response than transformers, but they're noisier. With the MA100, there was more than 20 dB difference in s/n between minimum and maximum, and noise increased almost linearly with the control setting. Using the Beachtek adapter with a low-level signal, there was about 16 dB more noise as the volume control moved from 0 to 100 percent, with most of the noise occurring with the control above 50 percent. With MIC ATT turned on, there was a 5 dB noise difference in the first 50 percent, and only 1 dB additional noise as you approached 100 percent. We suspect Canon used a lot of negative feedback with the mic attenuator, which is a good thing.

The Canon XL1's performance was typical of a well-designed but older camera.

Canon XL1 Noise and Distortion
Input Config.	Canon MA100		Beachtek DXA-4c		Audio 1 (rca)
Input Level	-35 dBu	-50 dBu	-35 dBu	-50 dBu	-30 dBV
THD + N	0.3%	1.2%	.24%	2.5%	.23%
S/N	54 dB	41 dB	54 dB	50 dB	61 dB

Canon GL2

Although the $2999 Canon GL2 is a lower-cost camera than the XL1, it's also a newer design (see the review in Oct. '02 DV). We suspect that some of its performance improvements over the XL1 are also seen in the XL1S. We tested the GL2 with Canon's $250 MA300 active XLR adapter.The relatively high 4 percent distortion for low-output mics (-50 dBu)-the worst we measured in any camera-is offset by the high s/n ratio. Canon probably used a little circuit trickery to trade one for the other. The performance with medium output mics (-35 dBu) was considerably better at 0.5 percent.Response was down 2 dB at 90 Hz and 9 kHz, and down 7 dB at 15 kHz. Aliasing was never worse than -30 dB, which was somewhat better than the XL1. Both of these factors suggest the GL2 has more modern and conservative filter design, although the camera doesn't seem to use oversampling.

Canon's GL2 has better audio performance than the older but more expensive XL1.

Canon GL2 Noise and Distortion
Input Level	-35 dBu	-50 dBu
THD + N	0.5%	4%
S/N	69 dB	60 dB

JVC GY-DV300U

We tested JVC's new $3495 GY-DV300U (see the review on page 59) through its built-in XLR input. JVC (http://pro.jvc.com) was the only manufacturer that disclosed even partial specs on its Web site. According to the site, the GY-DV300U has a frequency response of 20 Hz to 20 kHz. The manual quotes a response of 50 Hz to 20 kHz, but since both sets of numbers are meaningless without a decibel deviation number, feel free to take your pick. We were able to achieve the 50 Hz to 20 kHz range within a 4 dB variation-impressive for any prosumer camera-so there's no reason for JVC not to publish the full story.One other number turned out to be important, if you record with Automatic Level Control (ALC). The mic input level is specified at -55 dBu, much lower than most mics used in DV productions. ALC works fine at that volume. But it distorts with a signal of -35 dBu, typical of many shooting situations. We don't recommend using ALC for dialog with any camera; but if your situation requires it and you're using a DV300, you'll probably need an external attenuator. On the other hand, the camera's manual volume control was able to accommodate higher-level mics quite nicely.Response was down 2 dB at 67 Hz and 14 kHz, -3 dB at 50 Hz and 16 kHz, and -4 dB at 20 kHz. Aliasing distortion was similar to the XL1's under 12.5 kHz, although the DV300's extended high-end response naturally ex-posed more aliasing at very high frequencies.

The JVC GY-DV300U had impressive frequency-response performance.

JVC GY-DV300U Noise and Distortion
Input Level	-35 dBu	-50 dBu
THD + N	0.2%	0.15%
S/N	66dB	65dB

Sony DSR-PD150

The first Sony PD150 units ($4400) acquired a reputation for poor s/n. Eventually Sony developed a fix, which was built into cameras made after July 2000 and can be retrofitted onto older units through Sony's service center. We used a recent camera and ran the serial number by Sony to verify that it had the latest s/n improvements.We performed our tests with the MIC NR menu setting on and again with it off. The setting didn't make much difference. Sony's press representative couldn't put us in touch with anyone who could explain how the circuit works. As near as we could determine, MIC NR is a simple noise gate in an early stage of the preamp. We recommend leaving MIC NR off. You can do a better job of noise reduction in post.Response was down 2 dB at 160 Hz and 9 kHz, and -5 dB at 15 kHz. Aliasing was similar to the XL1's. The volume control was responsible for 1 dB of s/n change between 0 and 50 percent for low-level mics, with 8 dB more as it approached 100 percent. Mid- and line-level signals had no change in s/n up to 50 percent, and got 5 dB noisier above 50 percent.

The Sony DSR-PD150's performance was disappointing for a professional camera.

Sony DSR-PD150 Noise and Distortion
Input Level	-35 dBu	-50 dBu	Line
THD + N	0.3%	0.5%	0.3%
S/N	58 dB	52 dB	58 dB

Sony DSR-570WSL

The Sony DSR-570WSL is a newer professional camera with 2/3-inch CCDs that can record widescreen images and works with CCU-based studio setups, as well as field DVCAM recording. It lists for $17,200, about four times the PD150's price. As you might expect, its audio performance is also higher end.Sony recommends -20 dBFS nominal operating level-the usual standard for professional broadcast-but we took our readings at -12 dBFS so the results could be compared with other cameras.Frequency response was down 2 dB at 50 Hz and 15 kHz. Aliasing was single tones around -30 dB in the worst cases, which is considerably better than the prosumer cameras'.

The Sony DSR-570WSL, the most expensive camera in our test, had the best audio performance.

Sony DSR-570WSL Noise and Distortion
Input Level	-35 dBu	-50 dBu	Line
THD + N	0.1%	0.15%	0.1%
S/N	73 dB	60 dB	76 dB

Tascam DA-P1 Portable DAT Recorder

The venerable $2199 Tascam DA-P1 DAT recorder (www.tascam.com) is often suggested as an alternative to in-camera recording. Although it's an older 1994 design, doesn't support timecode, and is not up to current pro audio standards, the DA-P1's audio is a marked improvement over that in prosumer cameras.Our figures were from the DA-P1's digital output. There was a 14 dB decrease in s/n through the analog outputs. That's an important consideration, if you're comparing FireWire connected DV with a DAT that's being digitized in a sound card. Frequency response was down 2 dB at 20 Hz and 20 kHz, through both analog and digital outputs. Aliasing was -20 dB in the worst cases-not bad performance considering the extended response.

Tascam's DA-P1 DAT recorder was first introduced eight years ago. But it still offers better audio performance than any camera tested in this article.

Tascam DA-P1 DAT Noise and Distortion
Input Level	-35 dBu	-50 dBu	Line+4 dBu (xlr)	Line-10 dBV (rca)
THD + N	0.6%	0.6%	0.6%	0.6%
S/N	81 dB	70 dB	79 dB	77 dB

Sony MZR-37 Pocket MiniDisc

The MZR-37 Pocket MiniDisc is a consumer toy that cost less than $300 at Circuit City three years ago (and was featured in my Dec. '00 Audio Solutions column). Its performance surprised us.Our figures were measured using the MiniDisc's analog line output. We were able to get 3 dB better s/n and cut the distortion down to 0.02 percent by playing the disc through a nonportable deck with digital outputs.Response was down 2 dB at 20 Hz and 16 kHz. Aliasing was comparable to the prosumer cameras. Although MiniDisc uses ATRAC lossy compression, the algorithm is as benign as the highest-quality MP3 and shouldn't be audible with production sound.

This three-year-old consumer Sony MZR-37 MiniDisc recorder clearly outperformed all cameras except the $17,200 DSR-570WSL, and it bettered that camera in some tests.

Sony MZR-37 MiniDisc Noise and Distortion
Input Level	-35 dBu	-50 dBu	-10 dBV
THD + N	0.05%	0.05%	0.05%
S/N	72 dB	72 dB	80 dB

Getting the best sound

The charts on page 26 compare the cameras and recorders in four areas: Total Harmonic Distortion + Noise (THD+N), signal-to-noise ratio (s/n), effect of volume control on noise, and frequency response. We didn't plot every camera's frequency response, but we omitted ones that were similar. With this data and a little electronic common sense, we can make recommendations for getting the best sound from any prosumer camera.Avoid the camera's internal preamp as much as possible.In most cameras, you can't avoid the preamp entirely. But by keeping MIC ATT turned on and the volume control at 50 percent or below, you can keep its noise to a minimum. Many of the electret microphones used in DV production have enough output to work this way, when used at a proper miking distance. If your mic is less sensitive, get an external preamp or mixer to boost the signal. The preamp or mixer will also save you the cost of a separate XLR balancing adapter. Make sure the external device is calibrated so its 0 VU LED matches -12 dBFS on the camera (this may require an attenuator). In-line XLR attenuators are available from broadcast suppliers, and some DV dealers offer custom cables with that function built in. There's information about building your own attenuator at www.dplay.com/dv.In the Canon XL1, you apparently can avoid the internal preamp by using the RCA inputs with an external preamp or mixer. You'll probably need an attenuator for this as well. We found the XL1 seemed happiest when those jacks were getting about -30 dBV.The JVC DV300 doesn't have a MIC ATT setting, but its manual volume control accommodates both low- and high-level mics with a minimum of noise. Line-level signals distort in this camera, so you'll need an attenuator if you're using a line-level mixer.Turn off ALC and any microphone noise reduction option.Vendors sometimes claim ALC reduces electronic noise, but ALC also raises apparent background noise in most miking situations. The MIC NR setting on the PD150 had very little effect on performance. Both background and electronic noise can be considerably reduced in post, if you used good mic and recording technique. Noise reduction algorithms work best on tracks with little echo and a good ratio of dialog to background.Carefully monitor recording levels.The recommended -12 dBFS prosumer recording level doesn't have much margin for error. Sudden shouts or unexpected noises can cause digital distortion that can't be fixed in post. But if you record much softer than -12 dBFS, some dialog will be very close to the noise floor. So it's essential with these cameras to set levels carefully and monitor them during the shoot.One professional trick is to split a single mic's output so it is recorded on both camera channels, with one channel set about 6 dB lower than the other. In post you'll use the hotter channel, with less noise, most of the time. But you can switch to the quieter one if the hot channel gets close to distortion.Don't sweat the aliasing in most situations.The aliasing distortion in most of these cameras would be unacceptable in pro recording equipment; it's the kind of digital grunge that made sophisticated listeners reject early audio CDs. Fortunately, human speech and many forms of acoustic music aren't very strong in the high frequencies that trigger aliasing.Consider double system for theatrical or music-oriented projects.The previous tips will give you a track good enough for most broadcast and documentary projects. But portable DAT and even low-cost MiniDisc can give you a much better track. Although these media generally lack timecode, the decks themselves are inherently speed-stable. Use conventional film techniques to slate and manage the dual media-we recommend head and tail slates-and send a mic signal to the camera as well for sync reference and backup. Everything should stay in sync for at least 15 minutes. If you need much longer than that and can't use a cutaway while you trim picture to resync with sound, use a professional recorder that can share black burst with the camera.You'll get the best double-system results if you digitally bring the audio into the computer. If your NLE doesn't have digital inputs, it may be worth having a sound studio convert the field recordings to data files and bring them into your computer via CD-ROM.CAVEATSWe tested only one of each of these cameras, and it's possible, although unlikely, that we got a lemon. But this kind of equipment tends to be consistent-either it works as designed, or it doesn't work at all-and we got our test units from well-maintained rental and private collections. Manufacturers who feel they got short shrift are welcome to submit other units for analysis; we'll publish corrections if our tests show significantly different results.

Jay Rose is a Clio- and Emmy-winning sound designer with an abiding interest in the technical side of things. He delves into this topic and others at greater depth in the second edition of his book Producing Great Sound for Digital Video (CMP Books, 2002), due this December.Dan Rose is studio maintenance engineer and chief operator for NPR powerhouse WBUR, and is a member of the executive board of the Audio Engineering Society's Boston section. He's worked with Jay Rose on various audio projects over the past six years.Jay Rose and Dan Rose would like to thank Andrew Willis of Rule Broadcast Systems in Boston (www.rule.com), who provided most of the cameras, and Boston radio station WBUR (www.wbur.org) for some of the test equipment.