想知道RAM数据空间的差异

这意味着你的代码可能是好的。作为经验法则，优化在坏的斜率代码上工作得很好，但是它们不能对代码进行大量的补充。当然也有例外。



      以下为原文

   


This means you code is probably good. As a rule of thumb, optimizations work very well on bad sloppy code, but they cannot add much to the code which is already good. There are exceptions, of course ...

如果选择适当的系数，就不需要浮点。例如，几乎相同，但1／1024有利于整数的使用，与0.0.1相比，这会产生可怕的舍入，因此最好先将FY值存储乘以1024，这在简化之后产生toor，这是可以进一步简化或替换的。这是非常容易计算的分区，当你需要得到你的FY值，你也可以添加适当的舍入到您的班次。



      以下为原文

   


You do not need floating point if you choose appropriate coefficients. For example:
 
F_value=var_read*0.001+F_value*0.999;
 
is nearly identical to
 
F_value = (var_read/1024) + (F_value*1023/1024);
 
but 1/1024 facilitates the use of integers compared to 0.001.
 
However, this produces horrible rounding, so it is better to store F_value pre-multiplied by 1024, which yields to
 
F1024_value = (var_read/1024 + (F_value*1023/1024))*1024;
 
or after some simplification
 
F1024_value = var_read + (F_value*1023);
 
which is the same as
 
F1024_value = var_read + (F1024_value*1023/1024);
 
Which can be further simplified to
 
F1024_value += var_read - F1024_value/1024)
 
Or, replacing the division by shift
 
F1024_value += var_read - (F1024_value >> 10);
 
This is extremely easy to calculate, and when you need to get your F_value, you do
 
F_value = F1024_value >> 10;
 
You can also add proper rounding to your shifts.

这很有意思。你能告诉我数据类型是不是可以把一些预乘的结果保存在查找表中进行快速计算？



      以下为原文

   
 
That's very interesting once again.
Can you kindly tell me the data types for

F_value;
F1024_value;
var_read;
 
Is it possible to save some pre-multiplied results in a lookup table for fast calculations?

我猜想，所有这些变量都是未签名的16位整数。编辑：我错了，见下文。



      以下为原文

    I would hazard a guess that all those variables are unsigned 16 bit integers.
edit: and I'd be wrong, see below.
 
 

FY-Valk和ValuRead必须能够保持ADC值。如果你的ADC是10位，那么16位整数就可以了。F1024Y值必须是10位宽，10位+ 10位＝20位，所以32位整数就可以了。你所做的就是：不需要任何乘法。



      以下为原文

   
 
F_value and var_read must be able to hold the ADC value. If your ADC is 10-bit then 16-bit integers will do. F1024_value must be 10 bits wide, 10-bit + 10-bit = 20-bit, so 32-bit integers will be Ok.
 



All you do is this:
 
F1024_value += var_read - (F1024_value >> 10);
 
There's no need for any multiplications.
 
 

你能解释一下转变价值是什么意思吗？我必须说，这是一个非常漂亮和深思熟虑的数学编码方法。我正在读上面的帖子中提到的DSP指南手册，但是我在这里找不到这种简化的编码。这是你自己设计的威廉希尔官方网站 还是来自某本书？谢谢分享。



      以下为原文

   
Can you please explain what it means to round a shifted value?
 



I must say this is something very beautiful and deeply thought out way of coding the maths.
I am going through the dspguide book that is mentioned in a post above but this level of simplified coding i cannot even find there.
Is it your own devised technique or is it from some book?
Thanks for sharing it.
 

当你做浮点除法时，你得到分数：整数除法（或移位仿真器）只是砍掉分数：但是1.75比2接近2，所以最好是舍入而不是斩波。要做到这一点，你需要在操作之前把除数的一半加到股利上：在你的情况下：不。我真的记不起我什么时候开始使用这些东西了。那是很久以前的事了。



      以下为原文

   
 
When you do floating point division, you get fractions:
 
7.0/4 = 1.75
 
Integer division (or shift emulating it) just chops the fraction off:
 
7/4 = 7 >> 2 = 1
 
But 1.75 is close to 2 than to 1, so it is better to have rounding instead of chopping. To do this, you need to add half of the divisor to the dividend prior to the operation:
 
(7+2)/4 = (7+2) >> 2 = 2
 
In your case:
 
F1024_value += var_read - ((F1024_value + 512) >> 10);
 



No. Not from a book. I really cannot remember when I started using these. It was a long time ago.

准确度只取决于计算的算法是多么干净。使用AN575的引用，它有一个非常好的（古）算法。从该文中，求和使用248个周期，乘以574。所以总共是1396个周期，加上调用和数据的移动，所以大约1500个循环最大值。这对代码来说是可以接受的吗？编辑：我同意，使用整数移位的实现要好得多，更干净和更快，如果你接受平滑的因素限制为2 ^ NSO，我不知道这对PIC会不会很困难，我真的无法理解你的意思是什么…关于滤波器比较，我确信DSP书在过滤器之间有许多比较。而且总是有Excel，这是一块蛋糕来模拟所有这些东西。



      以下为原文

   

Accuracy depends only on how clean the algorithm that does the calculation is. Use the reference from AN575, it has a very good (ancient) algorithm. From that paper, sum uses 248 cycles, multiply uses 574. So total would be 1396 cycles, plus call and movements of data, so around 1500 cycles max. Is this acceptable for your code?
Edit: and I agree, the implementation using the shifts in integers is much better, cleaner and faster, if you accept the restriction of factors for the smoothing as being 2^n

So I don't know if it will be hard for the PIC, indeed I could not understand what you meant with hard ...
About filter comparison, I'm sure the DSP book has many comparisons between the filters. And there is always excel, were it is a piece of cake to simulate all this stuff.

好威廉希尔官方网站 ，我总是检查携带



      以下为原文

   

Nice technique, I was always checking carry

...
RRF myVar, f
BTFSC STATUS, C
INCF myVar, f
...