| ??? 03/12/09 23:45 Read: times Msg Score: 0 +1 Informative -1 Overrated |
#163397 - Give the customer value while making money Responding to: ???'s previous message |
Maybe I'm just tired because I have spent a couple of hours on planes. But I feel a need for a long relaxing vacation. But I think Richard should plan one too, to reduce the stress levels.
On the question how you can adjust the light intensity without wasting four or seven or a gazillion driver outputs for each pixel, I write: Per Westermark said:
You either use PWM and modulate the that 1ms of glory time for the scanline. and Per Westermark said:
All in a nice PWM module or using the timer capture/compare or something like that? And the assumption is that you have to go for a more complex processor: Richard Erlacher said:
PWM can be used for that, can't it? If you have PWM available in your MCU, or if you can justify it as part of the selection criteria, then yes, by all means, it should be considered. Yes, it is very very much considered. And possible to do with timers and interrupts or compare/match registers. Richard Erlacher said:
I'm not convinced that PWM as a voltage control is the answer Who talked about voltage control? Just energy control. About size of components: Richard Erlacher said:
Let's see ... a 16-pin SSOP's and a 20-pin soic vs a PLCC-44? Those two are not smaller than the PLCC, but not much larger either. Further, the ssop-16 can be put under (on the opposite side of the board) the SOIC-20 if necessary. I don't think packaging will be a problem, considering the 10:1 price difference. If necessary, that will easily pay for another PCB to economize on profile No, the ssop16 can not - but must - be put under (on the opposite side of) the board. The top side is already filled with the diodes. One thing discussed here is that when the diodes are small, you may have troubles to fit the driver chips for a fully DC sign even with 16-output combined shift registers/drivers. Going discrete would be totally, absolutely, fully impossible. For a multiplexed sign on the other hand, it doesn't matter if you can use a 44-pin PLCC in the same board space as a 20-pin SSOP since the limiting factor is amount of power loss allowed in the chip. In the end, you can't really go higher than 8-output chips right now. But even in a multiplexed solution, the number of components or solder points (and error source) would increase a lot if a commercial display isn't built with integrated shift registers/latchers/drivers. So packaging do matter for both multiplexed and DC-driven signs. Extra PCB to economize on profile? Remember that if you put two PCB above each other, you will have a bit of a problem with the cooling of any components on the inside of this sandwich. Since the diodes are obviously on the outside (the front) and since the reason for using two PCB is that the outside back is already full of driver electronics, the components on this inside would then have to be hot driver electronics, and we all know that the current they can support directly follows from the ambient temperature they are used in. If that extra PCB requires active cooling your MTBF will be affected. Richard Erlacher said:
Yes, but, unless you can find a more efficient LED, the power has to increase if the light output must increase. It's all energy ... if you need more light, you have to put in more electricity. 1 LED => 2V for LED + x V for constant-current regulator to regulate with. 2 LED => 2x2V for LEDs + x+eps V for constant-current regulator to regulate with, where eps < x or normally << x. 4 LED => 4*2V for LEDs + ... The energy loss in the constant-current regulator will decrease compared to the energy sent to the diodes when you drive the diodes in series. Four diodes in series do not require four times the voltage to regulate with in the cc driver. The required eps in this case can often be almost zero. And everyone (?) knows that if you have to transfer 1kW of power from point A to point B, then you will lose less power in the cable if you send it as 1kV 1A, than if you send it as 1V 1kA (unless you are willing to spend a huge amount of money for the transfer media. The vehicle industry looks at 48V systems in cars because of the cost of all the copper wiring in a 12V system. All traces in the PCB are resistors. The transistors in the row drive have a voltage loss. The PSU has efficiency figures. All these steps one by one all adds up to the same result: A display with multiple diodes in series in each pixel will be more power-efficient than a display with the diodes in parallell. And controlling he "on" time of the diodes will efficiently adjust their intensity in a huge number of intensity steps. Allowing a huge number of intensity steps by turning on a varying number of diodes would either require a huge number of diodes, or that you design the system so that the diodes are driven by n, 2n, 4n, 8n, ... current units so you can get 75% intensity by lighing the 8n + 4n diode and 87.5% by lighting the 8n + 4n + 2n diode... Does efficiency matter? If a text-only display, you normally have quite few LED lit at a time. In a billboard sign you may just as well display black text on a lit background, so you can have a very high average of diodes lit. Say 30%. Say the panel has 640x480x4 (monochrome). Say you drive 10mA/LED @ 2V (no need for max intensity). That gives 3686 amps. Say 2V for diodes and 1V for control = 11kW. Say you drive your diodes in series. 0.25 as many driver outputs. Now 8V for diodes and about 1V for control. 922 amps and 9V gives 8.3kW. If the customer finds that the display consuming 8.3kW on average looks visually more pleasing since the visual diode spacing doesn't vary with the environmental light - do you think they buy the "cheap" display needing 8.3kW, or do they buy the much more expensive display that looks worse and requires 11kW? Note that the above was just an example, without aiming for a specific driver chip so a real-life solution can probably reduce the difference between the two designs by 30-40% by letting the single-diode solution run with a lower regulation margin than the solution with all diodes in series. But it does show that power matters. And the above figures did not take into account the difference in losses in the cables and PCB wires and the efficiency figures for a PSU supplying 3V or one supplying 9V. Richard Erlacher said:
If I need a sign that has to be readable from 500 meters in daylight, I need a way to make it bright enough and to make the pixels large and bright enough to be visible. If there's a way to do that without increasing the LED count, driver count, power, etc, that's wonderful, but they must all be traded off in order to attain an optimal solution. The definition of optimal can vary widely, though. The way to do it without increasing the driver count is obivous - run them in series. The way to do it without increasing the power loss in the driver is obvious - run them in series. Richard Erlacher said:
If you have PWM available in your MCU, or if you can justify it as part of the selection criteria, then yes, by all means, it should be considered. 1) No need for a hardware PWM - software will do just fine. 2) The cost of hardware PWM in the processor would probably represent the cost of a single-digit number of diodes. PWM hardware isn't exotic. Richard Erlacher said:
I don't think packaging will be a problem, considering the 10:1 price difference. What 10:1 price difference? And when does a price difference matter if the cheaper solution doesn't solve the problem? It is irrelevant what discrete components costs if you can't fit them on the available space. Richard Erlacher said:
Let's see ... the current goes into the LED here ... and it comes out of it there ... through the CC current sink ... isn't that the case? LED forward voltage is the same ... How do you get MORE current to flow through the LED without having it flow through the current sink? If the current sink is constant-current, how would you increase the power? I must have missed something here. We already had a couple of posts mentioning the magic word: Series. You do not try to put more current through them. You try to let the cc driver supply current to multiple diodes in series. More power to the diodes but not more current through the driver and a small increase in the voltage drop required by the CC driver. Richard Erlacher said:
If you're just controlling duty cycle, and subsequently integrating the waveform, then it doesn't matter whether you have hardware or firmware control over it. In fact, all you need to do is shorten/lengthen the ON time. You certainly don't need dedicated PWM hardware. Correct. Requiring dedicated PWM hardware was your assumption. And using multiple fully on/off diodes instead of varying the duty cycle seems to be your recommended solution. Richard Erlacher said:
Whether you connect them in series or in parallel depends on your positive supply. Clearly, if you have multiple LED's per pixel and want to control them individually, you have to have correspondingly multiple drivers per pixel. Otherwise it matters not whether they're in series or in parallel. Correct. IF you want to drive them individually. You want to do it. But you will have troubles finding large numbers of sign manufacturers who wants this. A huge sign takes lots of power. Individual control has worse efficiency. And takes more components. And takes more processor capacity = time. Lots of reasons why the people who build signs do not see any advantage with this but do see big disadvantages. Richard Erlacher said:
3:1 for higher integration is, at least, worth a look, while 10:1 generally is not Relevance of price difference between solutions has to be scaled with their quality factor. In this case, the discrete solution may have a perceived quality factor of zero because of form factor, MTBF or cooling requirements. Richard Erlacher said:
In "normal" ambient lighting, you need only one pixel to be on at a time in order to ensure visibility, or, each one to be ON 1/4 of the time, which you can make to look reasonable by multiplexing the four LED's. Sometimes it requires two at a time. On the other hand, at 3 in the afternoon, the sun shines through the trees, window, or whatever, and makes the sign hard to read unless you really increase the illumination, so, instead of driving one LED at a time, and having to dim the room lighting, you drive all four. Would you prefer to buy a display with black and four light levels and very visible change between the levels? And the possibility of flickering if you have too little filter time + hysterese in the light sensor? Or would you prefer to buy a way cheaper display with less power consumtion for the same maximum display intensity and that will change intensity with almost invisible intensity steps? Richard Erlacher said:
The added benefit is that, with multiple LED's and multiple drivers, the drivers don't overheat, the LED's are still driven within their limits, etc. But the CC driver doesn't really care if it is driving one 4V LED at 100mA, or two 2V LED in series at 100mA so long as you adjust the input voltage. All it see is a current of 100mA and the voltage drop required to regulate the current to 100mA. The next factor is that if you run 1 to 4 LED, then at low light, one output will run at full power while three outputs will run at zero power. With the four LED in series you can request 25% LED intensity with a single output running at 100mA peak, but only 25mA avg which means that all output transistors in the driver chip can work at similar load. Richard Erlacher said:
It's simply a matter of using more of what you already use, rather than looking for something "better." What is wrong with looking for something "better", when something "better" happens to actually be better and cheaper? Richard Erlacher said:
What processing power are you hoping to save? What would a self-contained display do aside from display the characters, monitor ambient light, and receive new buffer contents? It could, I suppose, manage its power supply ...
This MCU has little to do other than to operate the scrolling display. It uses 1280 pixels to display two lines (?) of 80 characters (The O/P said it was two lines of English in Roman alphabet.) He uses up 80 memory locations just for his text, and then has to manage the rest of the display and scrolling operation. You think that a manufacturer designs a display board specifically for a 80x16 display? And then design another board for a 160x32 display. And a third design for a 240x8 display? The display board beside my computer runs 960x64 pix. If I need to control four outputs for each pixel then I need four times as much image memory for the bitmaps. And I obviously need to generate the data for four times as large bitmaps. And I need four times as high clock frequency when sending out data to shift registers or emit data on four times as many chains or let the shifting take four times as long. And the display board may not just "display". It may have to spend time communicating with something or computing something. Possibly running a full weather station to let you know temperature, wind, ... at the top of the downhill slope. You can almost always find faster processors. Or you can add more processors. But the fact is that wastefully adding requirements for extra outputs removes one degree of freedom. Freedom that could have been used for solving a problem the customer actually can make use of. Maybe doing real-time video instead of just scrolling horisontal text. Your suggestions seems to end up costing the customer more. More board space for control logic. More processing power needed to drive the image. More power loss feeding the sign. Do you think that sign manufacturers don't know what they are doing? A good designer should try to reduce the production cost and maitainance costs so that the seller makes more money and at the same time give the customer an additional value so that they select your product instead the competitors. Telling that your product consumes 20% more energy because that mamoth sign with 640x480*4*3 (full-color with four diodes times three colors/pixel) decides to run the diodes individually instead of in series doesn't help. To sell them a 1280x960x3 full-color display for the same price would probably be an easier deal. Because that is basically what you are suggesting: A high-res sign sold as (and forcibly limited to be) a low-res sign with multiple diodes/pixel. How much would you pay for a TV with a HD panel controlled in a way that it gives SD resolution and allow 5 levels of R, G and B for a total of dithered 125 colors? How much would you pay for a TV with the same size display, but built with a SD panel and where you instead have intensity control on the pixels? |
