Sliding into this legendary thread. Lack of knowledge on opamps and filters so let me contribute.
OPAMPSFirst rule is you want gain bandwidth product of opamp to be at least [10 x gain x max frequency in signal]. If we say [Standard Definition aka 15k aka 480i/576i + 240p/288p] video caps at 5 MHz to be extra generous then we need 10 x 2 x 5 = 100 MHz to the -3 dB point you see on a Bode plot. Datasheets usually list the bandwidth but have to infer if it's at unity gain or the fixed gain in video amps. 80 MHz at unity gain is effectively 40 MHz with gain of 2 - too slow when we want 50+ MHz. Basic reason for min bandwidth is the operating rate of the IC is itself a lowpass filter (LPF) because it can't handle signals faster than it can run. We lose 50% power at 3 dB so the 10x margin is 1 decade of safety. Incidentally, +6 dB means doubling the signal's power for a voltage gain of 2.
Slew rate is how fast the opamp draws the output. Math is easy enough, if we need on the output 1.4V then takes 1.4V / [slew rate V/us] = time in microseconds to draw the rise or fall time. In practice, takes longer for a few reasons but can use that to sort opamps for the minimum slew rate you need and then analyze datasheet graphs of rise and fall times. If sinusoidal signal instead of RGB square wave then min slew rate you need is [2 x pi x Vp * max frequency in signal]. Have a margin of safety.
The reality is THS7316 is nowhere near fast enough to amp analog video at unity gain bandwidth of 36 MHz without distortion. THS7314 hides behind its mandatory LPF to not even list its bandwidth in datasheet but can zoom in on page 9 to see it's slow as balls as it starts attenuating before 5 MHz:
Neither THS7316 nor THS7314 should have ever been used in our scene.THS7376 and THS7374, those are fast enough for analog video. Can debate on which 'fast enough' amp is better than another due to noise specs, differential gain/phase error, higher speed or faster settling time. I think LT6550 for instance is better than those if you don't need the LPF and you don't mind paying $4-5 vs $1. At luxury $6-7 tier, you can use LT6557 or OPA3692 with insane 2000V/us slew rates. Extra speed does reduce the gain error with diminishing returns and read datasheet for notes on stability. Speed can be too much of a good thing.
If you want to amp luma (Y) with its negative sync voltage without cutting it off, or allow DC coupling of RGB, I'd suggest looking at ISL59837 that has a built-in charge pump for negative voltage. Renesas also makes a superior sync stripper compared to LM1881.
One more thing. Quad channel opamps like THS7376 and THS7374 where you don't use the 4th, you need to
terminate it correctly or you add noise and generate more heat. Is a design flaw this scene has had for years that I'd like to see fixed.
--------------------------------------------------------------------------------------------------------------------------------
FILTERSFirst, you don't need a filter playing on (analog) CRT. Is a waste and just adds distortion. Noise at 6+ MHz that a 8.5-9MHz LPF attacks is invisible to 480i/576i SD CRT. Most people know but I prefer to have all basic info one one place.
You definitely want a filter before analog to digital conversion to prevent aliasing for HDMI or otherwise scaling from interlaced to progressive. Framemeister, OSSC, RetroTINK 5X Pro have them but not sure about other RetroT scalers or capture cards.
Filters, chaining two independent LPF or HPF together significantly reduces the cutoff frequency and should never be done. Chaining two 2nd orders to make a 4th order filter with worse characteristics than designing a 4th upfront is classic lab exercise. Each opamp can have 1 or 2 orders on it so 4th order requires 2 opamps. Odd orders, sometimes you'll see 1st order simple RC passive filter chained on at the start or end to save an opamp. Passive filters beyond 2nd order aren't useful in practice because their max gain is 1 and have limited passband compared to active that applies power to preserve a constant gain that can be >= 1. Video amps have gain of 2 to compensate for 75 ohm voltage divider on the video processor.
What is an order? Each capacitor and inductor in the filter that doesn't get canceled out, adds 1 order. Can equivalently say that the order is the highest order in the differential equation of the circuit's time response or the highest degree in the polynomial of the denominator in the transfer function. Whew.
Each order increases the rate of attenuation by 20 dB/decade (good) but also increases the phase distortion and overshoot/ringing (bad). Classic tradeoff between better frequency response at cost of worse phase response. We all seem to use Butterworth filters that have balance of good frequency and phase responses. Chebyshev and Elliptic have superior (faster) frequency rolloff but worse phase responses. Can get away with a lesser order for same attenuation. Bessel, common in audio amps, has best phase response but slowest (worst) frequency response that audio world accepts to preserve sound quality.
Rate of change of phase delays is the group delay. What you want to minimize is the non-linear group delay that is shown on opamp datasheets. Effect is stretching and compressing the frequencies by having them arrive at the output at different times. Too much of that and video will be distorted. Linear group delay, is okay, is effectively lengthening the cable distance but too much and sync that isn't filtered with this added delay is, well, de-syncing. Can separately delay the sync with an all-pass filter or other method to match video's arrival time.
So why is SD lowpass filter cutoff of 8.5-9 MHz when signal doesn't go past 5 or 6 MHz? A brickwall filter that passes 5 MHz and completely cuts off 5.1 MHz is physically impossible. Got to start your attenuation past the end of your signal and accept not filtering most of the noise in that 5.1-8.0 MHz band. An unfortunate amount of electrical engineering is a balancing act between the ideal and reality. Need a 355pF capacitor? Doesn't exist, take your choice of 350 or 360.
Inductors aren't used in active filters because of how large and non-ideal they are compared to capacitors, not to mention they cost more. They are used in the form of (passive) ferrite beads that you see on PC power supplies as cheap LPFs and perhaps they could become a thing in our scene.
FAQ
Search
Register
Login
