Real Resolution of a Standard TV?

[Ted Springer]

So what is the real resolution an average NTSC television is capable of? I have heard that it is 512x384, but that really doesn't make much sense. Maybe in non-interlace mode (yes, NTSC can do non-interlace). But I was under the impression that it was either 640x480 or 720x480 or so.

[Mike Rehmus]

There are many answers to the question but the 'accurate' answer is 525 lines minus the lines times that are consumed by synch and other information such as SAP, etc. Analog televison doesn't really have a horizontal resolution, it has a luminance maximum frequency which can be converted into a maximum resolution.

What it normally comes down to is the quality of the televison receiver and what it can do.

Digital editors have to have a resolution figure and various 'standards' are out there. 640 by 480 is one, 768 by 572 is another. 720 by 480, 720 by whatever you want.

[Ted Springer]

I would assume that 525 lines minus the needed information would equate to about 480 lines, some of which is way past the safe area, correct?

[Rob Belics]

Yes. The extra 45 lines include the vertical blanking sync, horizontal sync, vitc, and other digital information sometimes.

The number of pixels on a line is limited only by the transmission circuits. Still, the ability to resolve them on a monitor or TV is the main issue as Mike said.

If you take the vertical timing as 29.97 per second and divide it by the number of horizontal lines (525) you get about 63.5 usec per line. Subtracting the horizontal blanking time, which includes color burst and horizontal sync, you wind up with, uh, I forgot. Let's say 60 usec (that's wrong). If you divide 60 usec by the number of pixels on a line, say 640, you'll get around 100ns. Which translates into about 10Mhz. This is luminance only.

(Note: I'm doing this in my head and it's been a while since I've had to do this so I'm sure, if you need it, someone can fix my numbers or correct assumptions I make below).

The FCC does not allow a 10Mhz bandwidth, so filters are applied to limit it. The actual luminance is closer to 384 or even 280(?).
The actual bandwidth allowed is 5Mhz or less.

But color burst is at 3.58 Mhz. So filters notch out that spot so it doesn't interfere with luminance, further limiting resolving ability.
Color info rides on the color burst but must stay within the allowed bandwidth. Therefore color info displayed on the screen is extremely poor; much less than the luminous resolution.

Like I said, I'm trying to recall this but I should be in the ballpark. It's sad to think I used to design circuits like this for a living. But I used to repair TOW missile systems, too, so, whatever.

[Helen Bach]

There are two resolution figures here, aren't there? The number of visible horizontal scan lines is pretty much fixed by the system (PAL/NTSC etc) as has already been written. The resolution along the line is highly variable from one television/monitor to the next. Look at some monitor specs for their resolution figures (the B&H online catalog is a handy collection of specs). 500 lines is good, 800 is exceptional, 300 is not unusual. These are measures of the performance of the television/monitor itself.

The resolution of the incoming signal is another factor, and Rob has covered that well.

Best,
Helen

[Mike Rehmus]

But between the so-called camera resolution and the monitor are a lot of slippery issues.

There are stated performance figures for the gear we use. Those values are usually better than what we actually experience in the real world.

Some camera's optical blocks are said to be capable of as much as 900 lines of horizontal resolution. OK, what does that mean? It depends.

Here is why: Conventional expressions of resolution take in to account, the ability to not only resolve details but to deliver them in a viewable form. To do that, there has to be some space between the detail elements or we are looking at a smooth surface. Like an uniform color matte that we might use as a background.

The claimed 900 lines of resolution are not really a real-world resolution value. Because the manufacturer's don't disclose how they arrive at the 900 line value, the measurement also cannot be reconstructed by a normal user.

Really useful resolution figures take into account the need for a wee little bit of space between detail elements. The measurement is called Line-pairs. A line and a space making up the pair and each of the same width. And it is further complicated by measuring how well the device can discriminate or display fully black and white lines side-by-side.

So, in theory, the 900 line camera can acually display just 450 line pairs. But wait, there is more! Can it really measure 450 line-pairs or is that a dream. The better the lens and optical block, the closer to reality it gets. The additional measurement that means a lot in this case is Modulation Transfer Function (MTF), a term you've all heard if you mess around with high-quality still camera lenses.

Approximately, MTF is a measurement of an optical devices capability to measure or display line pairs with a third dimension thrown in . . . luminance. Let's stick with black and white targets for this explanation. Assume each line is maximum white and each space is maximum black. For NTSC television in the U.S.A., that would be IRE values of 100 (white) and 7.5 (black).

Ok, if there is only one line pair in the field of view, say a picket and a space on a picket fence (It isn't finished yet) and they are sized and spaced to make a line pair it is not hard for us to image that pair as full white and full black. (except at the edges . . . more about that later)

Now our carpenter keeps adding pickets (line pairs) to the fence. Still no problem because if we can image one line-pair, we can image them all if we ignore the fact that lens systems tend to achieve poorer resolution at the edges of the field of view. We can, in this situation, generate a decent image that is composed of a bunch of white and black lines.

Now we are going to move the camera back and measure what happens to the image.

For a while, the line pairs get visually smaller in width but are still full white and black. As we get further away, however, the image of the pickets starts to change. The blacks start to become gray and so do the whites. But we can still determine the pickets are all there.

Now we get back to a point where the picket fence is nothing but a uniform gray blur. If the fence were tall and wide enough that it covered the entire field of view, we now only see a featureless gray field.

Somewhere during that pull-back we reached a point at which the fence no longer looked like a fence but sort of a series of dark and light bars. At some point in that continuum from clearly defined line-pairs to the gray field, we decide that if we move any further back, the image has no value to us. Lens manufacturer's used to say that was at where the pickets were about 70% of full white and black.

So at a MTF of 70, we've reached the resolution limit of the system. And that's sort of how the manufacturer's determine the resolution value that they advertise. Except the camcorder manufacturers sort of use double-speak. In the less expensive camcorders, they tell us the capabilities of the digital recording system. 500 to 540 lines (not line pairs) of horizontal resolution is what they normally claim. They don't say anything about the lens and optical block capabilities. But you can pretty much assure yourself that the lesser cameras are not going to deliver an image that can take advantage of the full recording resolution.

As soon as the camera head will accept changeable lens, the camera manufacturers (who don't make lenses for them) start claiming the max resolution of the optical block as delivered to the Composite Video BNC connector (which is not the same signal as the one that passes through the 'camera electronics'. Normally one finds the direct Optical Block output near the front of the camera and the video signal, after it has passed through the recording electronics, near the rear of the camera. I'm obviously talking about a pro camera here.

When we add in the effects of the lens on resolution, we will normally find that the lens reduces the resolution of the camera. Yup, we probably don't get 900 lines of scene resolution from the camera head signal. The PD150 likely does not record 540 lines of scene resolution although it is probably fairly close. The DV recording process certainly records 540 lines (270 line-pairs) of resolution but the lens/optical block combination very likely does not deliver that resolution to the recording circuitry. Certainly a $500 DV camera probably does not deliver a full resolution image to the recording system. I don't even know if the PD150 or the DSR-300, for example, are fully digital from the CCD all the way back to the recording system. Many cameras have some analog circuits right in back of the CCD. Unless the manufacturer states that the camera is fully digital, one just doesn't know. Not being fully digital means that the CCD signal isn't exactly in synch with the recording system and therefore the CCD output may not be fully sampled by the recording system. Very close but not exact.

And in any case, unless we just happened to align the projected image of the pickets exactly on every other CCD sensor, we could not achieve maximum resolution of the pickets. So no matter if the camera is all digital, the world is not. Isn't it great that we want to record people and other lumps of scene and not just picket fences?

So what does 900 lines of resolution mean? Or 800 or 300? Turns out that the camera manufacturer, the television studio and every one of us have no means of measuring the resolution of the image that is seen by our customers. The television is in their home. Their television may be the latest digital technology or it may be a 1953 RCA monochrome set. So all we can do is assure ourselves that what we place on the recording medium or what we transmit down the cable or to the antenna is good enough for our purpose, whatever that is. Then regardless of the signal degredation, we've done our best.

Our measuring tool is the waveform monitor and a good generator of test waveforms. (Color bars are only used to set up the display of the images, not to make resolution measurements in the studio.)

Remember Rob's 10 Mhz value? What that was derived from was the 100 nanoseconds (100 billionths of a second) that is allowed electronically by the circuitry. For simplicity's sake, lets call the the fastest speed at which the system can go from full-white to full-black. That's called 'Rise-Time' and is a Very Important Measurement. Please note that for the purists, the conversion I've just discussed is not quite as simple.

Signal Rise-time, an electronic waveform measurement perfomed on a waveform monitor, is transformed into Edge Detail when we look at a video display. You've seen a Luminance waveform if you've ever looked at a waveform monitor with a video signal present. If you sent the waveform monitor a real test signal, say a signal that starts out at one side of the line at 1,000 cycles (Hertz) and sweeps up in frequency to 10 Mhz as it gets to the other side, you can measure what rise-time your system (up to that point) is capable of delivering. Rise-time can be converted to an equivalent bandwidth and vice-versa. So you can measure rise-time and know what the bandwidth of the system is or measure bandwidth and know the rise-time (with one slight gotcha).

Rise-time / Bandwidth determine what amount of image detail that is delivered to the recording system, read from the recording system, or delivered to the transmitter or the cable system head-end modulator.

So where was I? Oh yea.

Now we get to the point where the color monitor or B&W monitor specs come in.

Cathode Ray Tubes (CRT), the 'picture tube' create an image from a red/green/blue (RGB) triplett group. Hit them all with an electron beam and you get a white pixel. The spatial frequency of the tripletts in the X & Y directions on the face of the CRT determine the upper resolution of the tube and the image it can display.

This is a mechanical limitation of the tube and cannot be changed no matter what signal you send the display system. So a 800 line monitor cannot show 800 line pairs, it can show 400 line pairs at some MTF (which they do not define or even talk about). But 800 line monitors generally have higher quality CRTs than 300 line monitors. So they can display higher resolution images . . . and they usually have more tripletts in the horizontal direction in their display tube. Their electronics are typically better and so are the input options.

Continued in the next post

[Mike Rehmus]

Monochrome monitor exception. The phosphor in a monochrome CRT is not deposited in spots on the tube face as is the case with many color CRTs. The size of the bits of phosphor determine the maximum resolution of the CRT. The way they deposit the phosphor on the face of the CRT is to mix it in water, pour it into the glass envelope of the CRT held face-down and allow the water evaporate. The result is a fairly random placement of phosphor granules. As you might imagine, most monochrome CRTs are capable of much higher resolutions than color CRTs. That is the major reason pro viewfinders are monochrome and why the camera control technicial and the camera operator in a television studio uses monochrome monitors for each camera in the studio. One on top of the camera and one at the camera control station.

So if we want to show a maximum resolution picket fence on a monitor, we cannot show any more line pairs than there are pixel tripletts divided by 2, right? Wrong. We probably can only get close in most cases. We take a 400 line-pair analog signal (composite or S-Video) and feed it into this 800 line monitor. But this is an analog signal and so is the monitor (up to a point). The CRT can only display 400 line-pairs if we hit each triplett right on the nose. But we have no way of assuring ourselves that this will occur. Why? The scanning electron beam that makes the triplett pair's glow can be misadjusted left or right relative to the screen. So it could hit part of one triplett and part of another. Or the analog signal could just be timed such that the beam is 'On' such that it covers part of two adjacent tripletts. (Note that this ignores the issue of the beam being tilted left to right and therefore sweeping more than one line of tripletts or that the vertical positioning can be off or all can be off at the same time.) Now you know why CRT systems are delicate and don't respond well to bumps and vibration.

But can we take advantage of a 800 line monitor? Yes, all things being equal, it will show a more detailed image and because it is intended for pro use, it has more stable circuitry and is all around a better display device for a pro. Pro monitors are (or should be) frequently mechanically and electrically aligned so they present as true a representation of the signal as possible. It should be noted that televisions sets are designed to correct out-of-spec signals so they deliver a good picture and are therefore not good for our purposes. (The design goal of the pro monitor is to show the true signal, not make it look good)

If we want to see 400 line pairs on a 800 line monitor, can we? Yes, if the monitor is a LCD and has a digital interface and the signal is always digital from start to finish. One of the reasons fully digital television should be pretty good viewing. Otherwise only if the moon is in the correct phase and we are very lucky.

BTW, if you wonder why we measure Rise-time instead of bandwidth, it is for this reason:

If you have looked into a pro viewfinder, you have noticed that when the camera is in focus, there is a thin white line around image details. This is because the viewfinder has a Peaking-circuit that distorts the video signal. If you feed a square-wave into the viewfinder, the circuitry puts a peak on the top of the vertically rising portion of the wave and a trough on the vertically falling wave. This causes the white line you see which is so valuable for determing optimum focus. Useful but it creates a bad picture for pure viewing. Were we to measure the bandwidth of the signal coming into the viewfinder vs the signal after the peaking circuit, you would measure a higher bandwidth inside than for the incoming signal. A single measurement after the peaking circuit would not tell you that the signal is distorted. But we want as accurate a representation of the scene as possible so we don't want signal distortion.

BTW, BTW, Most of our gear is quite capable of creating 'illegal' (better than) broadcast video. That's good, actually, because if it could only created to-spec video, then the actual delivered video would be much worse because of all the times the video undergoes some sort of conversion from the optical block to the home receiver.

A little more background:

If you've ever looked at a television resolution chart, the ones that have the converging black and white lines that are fan-shapped? Somewhere between the open end and where they converge, the lines merge into a gray. You decide where you cannot see any meaningful detail in the fan, read the number off to the side and that's how you measure the resolution of the entire system. Creation all the way to the viewer. Except we don't usually get to go into the homes of our viewers and probably would not be adjusting their sets if we did go there. How do we then measure what we create so the best possible image is viewed by the customer. (Best possible meaning within the constraints of their television.)

Is this all confusing? Yup. But important if we want to deliver the best possible signal. The above is not an education in television technology. A good source for some of that would be the Tektronix web site www.tektronix.com for info on television waveforms. The absolutely best place is to go apprentice or just volunteer to work in a studio for the folks who have to make it all work Right. Fortunately for most of us, the equipment we have will deliver acceptable image quality to our customers who are happy with VHS to DVD quality. But without attention to the details, we can never hope to deliver really good and clean video signals.

[Dan Uneken]

Wow!
I am going to print that and put hardcovers on it and put it with my other tech books.
Thanks very much!!

[Andre De Clercq]

Maybe some additions to Mike's posts
Resolution ""along the lines"in TV and video is being expressed as TVL/pph which means the max number of "discernable"lines on a horizontal length which equals the picture height. The theoretical 540 TVL limit for DV (720 hor pixels) results from this definition (720x3/4=540)
"Descernable"in TV, quite often means MTF levels at -10 to -20db (30 % or 10% as opposed to the 70% used for lens specs). CRT MTF is further strongly influenced by the contrast setting (spot blooming) and is therefore defined under low contrast conditions.
CCD resolution is a bit tricky and depends of course on the pixelcount, but also on the amount of optical aliasing that is being allowed. Before Mike's picket fences get gray they show strange repetitions (straddling). Pro cams apply optical anti aliasinfg filters to reduce these effects and get close to 540 TVL without too much of this aliasing effects . (b.t.w. Also color CRT and all displays with repetitive structures suffer from that kind of spatial aliasing) Consumer cams get their (low performance) optical anti aliasing äutomatically because of lens MTF limitations, diffraction effects and the higher pixelcounts (the latter at the expense of sensitivity and vertical streaking) .

[Rob Belics]

Man. This is the one thing I hate about talking to engineers and reading their reports.

I've already forgotten half of what you guys wrote but lens considerations don't matter when it gets down to transmission of the analog data. It all comes out analog eventually and, believe it or not, CCDs are actually analog devices along with their associated circuitry.

Maybe I'm saying the same thing Mike said but the reason a picket fence goes to gray is the frequency domain of the circuitry and won't matter if the lens is of good quality. HDTV, of course, will allow better resolving power of that picket fence due to wider bandwidth of the transmission system (along with compression).

If the rate of change (frequency) from black to white on a horizontal line exceeds this bandwidth, the system can only react part of the way, hence the gray appearance.

[Mike Rehmus]

You are correct, Rob, the lens isn't directly involved in the transmission and reception of the television signal.

It is, along with the camera, the other studio and editing equipment vitally important in creating program material that should start out as quite good quality if it is to survive with some semblance of quality when the resulting image hits the viewer's eyeballs.

So if we input a clean picket fence image and it comes out gray on the other end, then somewhere we have probably exceeded the bandwidth of the transmitter and/or receiver system.

Andre. Yes, if we start evaluating sampling frequencies, physical or electronic, it stirs in another level of considerations.

Always makes me curious in what the image quality differences are between on of the megapixel sensor cameras that map multiple CCD sensor sites through some processing to a single video pixel (like the Sony PC110) and the cameras that map one CCD sensor site to one video pixel (Like the PD150). And how the measured results might change as the scene changes.

Other than sensitivity issues, anyone have a 'read' on the technique and the measured results?

It was all so much simpler in the 80's when we started moving a one-line CCD array across the focal plane of a 35mm lens and calling it a digital camera! Took 3 scans to develop a color image and as long as nothing moved scan-to-scan and frame-to-frame, the image was outstanding. Also so much slower.

[Andre De Clercq]

Mike, the one to one mapping will not need to electrically lowpass filtering because there is synchronious sampling involved between the read data and the DV sampling rate . If the pixel count doesn't match up with the sample frame there is some extra filtering (convolution/ interpolation) involved. This filtering is independant from potential optical aliasing effects. The ideal CCD structure w.r.t. optical spatial aliasing would however have an infinite number of pixels.

[Rob Belics]

But originally we were talking about ntsc signals.

[Mike Rehmus]

Oh, NOW you bring us back to the topic! :-))

Actually the original question was about NTSC television set capability which is not necessarily about NTSC signals.

Truly, without measuring the TV, it is hard to give an accurate answer. The specs are all over the place and the age of the TV factors into it too.

I expect that a modern solid-state television has more resolution than I can deliver with VHS video. If the TV has S-Video inputs then I expect somewhere around 400 lines. If it has Component inputs, then I expect close to 500 lines.

As a rule of thumb, it works for me.

[Tim Frank]

How can the high-end TV's at say BestBuy say they have 625 lines of resolution or 600 then if it is not capable of displaying that many lines, would that be for something such as an S-Video input for a DVD player?