WEBVTT

00:00:20.310 --> 00:00:23.680
In this video will be talking
about the floating point

00:00:23.680 --> 00:00:27.050
notation, which is designed to
address the problem that we

00:00:27.050 --> 00:00:30.757
talked about in one of the
previous videos, where we noted

00:00:30.757 --> 00:00:34.801
that the radix point was jumping
around and I mentioned that it

00:00:34.801 --> 00:00:38.508
can cause a bit of a confusion
for the computer. Throughout

00:00:38.508 --> 00:00:42.552
this video I will be talking
about the 8 bit floating point

00:00:42.552 --> 00:00:46.596
notation which you will find out
very soon. It's a really small

00:00:46.596 --> 00:00:50.303
system, so it doesn't allow you
to work with huge numbers.

00:00:50.330 --> 00:00:55.159
However, it's convenient enough
to understand and work with on a

00:00:55.159 --> 00:00:59.988
paper, and you can just extend
the general rules to any

00:00:59.988 --> 00:01:04.378
systems. So let's see what is
this floating point notation

00:01:04.378 --> 00:01:09.207
looks like. Well, the floating
point notation is, 8 bits are

00:01:09.207 --> 00:01:14.475
broken up into three parts. The
first part is just one single

00:01:14.475 --> 00:01:20.182
digits telling you what is the
sign of your number. The next 3

00:01:20.182 --> 00:01:25.089
digits. We called the exponent,
which will tell the computer

00:01:25.089 --> 00:01:30.981
where to move the decimal point
from here and the last four

00:01:30.981 --> 00:01:36.052
digits. Called the mantissa. Now
the mantissa is the normalized

00:01:36.052 --> 00:01:40.032
version of your binary number
into the 0.122 format. We're

00:01:40.032 --> 00:01:44.808
going to go through it in in
details. Now. The exponent bit

00:01:44.808 --> 00:01:49.982
in this format is just three
bits, so you can only move your

00:01:49.982 --> 00:01:55.156
decimal point to three in to the
right or to the left, which

00:01:55.156 --> 00:01:59.932
straight away as it as you that
you can't really cover great

00:01:59.932 --> 00:02:01.922
ranges, but it's good enough.

00:02:02.760 --> 00:02:07.105
And the exponent is expressed as
a 3 bits 2's complement

00:02:07.105 --> 00:02:13.752
notation. So the sign bit if
we have a positive number be

00:02:13.752 --> 00:02:20.148
going to use zero if we have a
negative number, we're going to

00:02:20.148 --> 00:02:26.052
use one. As I said, the exponent
is 3 bits 2's complement.

00:02:27.060 --> 00:02:33.561
And the mantissa, which is
always 4 bits in this particular

00:02:33.561 --> 00:02:37.107
number, we're going to have it

00:02:37.107 --> 00:02:44.101
normalized. So when you
have got a negative

00:02:44.101 --> 00:02:49.099
number, something like
minus 3 1/2.

00:02:50.860 --> 00:02:55.624
How are you going to convert
this into an 8 bit floating

00:02:55.624 --> 00:02:57.609
point notation while step one

00:02:57.609 --> 00:02:58.403
will be?

00:02:59.410 --> 00:03:04.538
Identify.
The sign bit.

00:03:06.510 --> 00:03:12.321
That this is going to be the
easiest one for negative #0 for

00:03:12.321 --> 00:03:16.791
positive number. Step two will
be is convert into binary.

00:03:20.660 --> 00:03:23.260
And normalize

00:03:25.040 --> 00:03:27.364
To get to them on this so.

00:03:28.750 --> 00:03:32.045
Step

00:03:32.045 --> 00:03:38.940
3.
Is find the exponent.

00:03:39.610 --> 00:03:45.180
Which will be something
to do with the month. So

00:03:45.180 --> 00:03:49.636
how you normalize your
number and express it.

00:03:52.120 --> 00:03:53.869
As three bits.

00:03:55.460 --> 00:03:59.415
These complement notation and
the last step.

00:03:59.970 --> 00:04:04.227
Is just to pull everything
together and put it into the

00:04:04.227 --> 00:04:08.871
format of the sign followed by
three of the exponent digits and

00:04:08.871 --> 00:04:10.806
followed by four of the

00:04:10.806 --> 00:04:15.260
mantissa. OK, it sounds quite a
long and probably a little bit

00:04:15.260 --> 00:04:19.740
alien in theory, but I think
what we are going to help us if

00:04:19.740 --> 00:04:21.980
we go through a couple of simple

00:04:21.980 --> 00:04:25.095
examples. So let's look at minus

00:04:25.095 --> 00:04:27.029
one. And a quarter.

00:04:27.670 --> 00:04:31.900
OK sign bit going to
be one because this

00:04:31.900 --> 00:04:33.780
is a negative number.

00:04:36.040 --> 00:04:41.200
Not we're going to find the
mantissa for that first. What we

00:04:41.200 --> 00:04:45.930
need to do? We need to convert 1
1/4 into binary.

00:04:47.660 --> 00:04:51.170
So. Let's again bring

00:04:51.170 --> 00:04:57.870
in the. Place values Now this is
the radix point. Here is 124 and

00:04:57.870 --> 00:05:02.790
the rest of them. The whole
numbers and after comes up half

00:05:02.790 --> 00:05:07.710
a quarter on 8 etc etc. Now I've
chosen deliberately choosing a

00:05:07.710 --> 00:05:14.270
simple example so to write 1 1/4
in binary or we need to do is 1.

00:05:14.870 --> 00:05:19.814
Radix Point followed with 01. So
this is our binary number, but

00:05:19.814 --> 00:05:22.698
we need to normalize it into the

00:05:22.698 --> 00:05:26.903
mantissa. So the normalized
the normalized monthly start

00:05:26.903 --> 00:05:30.674
then will be 0.101. And
remember the amount Esther

00:05:30.674 --> 00:05:36.959
needs to be 4 bits, so I need
to put an extra 0 into here.

00:05:36.959 --> 00:05:41.568
Now what happens in this case?
Now I found my mantissa.

00:05:43.020 --> 00:05:47.700
I found mine on Monday so now,
but I still need my exponent.

00:05:48.390 --> 00:05:53.174
The exponent the easier way to
think about it is to try to

00:05:53.174 --> 00:05:56.854
think about the exponent from
the computer's point of view,

00:05:56.854 --> 00:05:58.694
because once the computer read

00:05:58.694 --> 00:06:03.340
the. Sign me and the exponent
wait, then the computer will

00:06:03.340 --> 00:06:08.215
know where to start from and how
many places to move that Radix

00:06:08.215 --> 00:06:12.715
point. Now from this radix point
here in them from the mantissa

00:06:12.715 --> 00:06:16.840
to get back to the original
number which remember was minus

00:06:16.840 --> 00:06:21.340
one and a quarter to get back to
the original number. The

00:06:21.340 --> 00:06:25.465
computer will need to move one
places into this direction and

00:06:25.465 --> 00:06:29.590
this is the positive direction.
So I'm moving one place too.

00:06:29.620 --> 00:06:34.050
The right that's a positive.
It's like multiplying by 10

00:06:34.050 --> 00:06:38.037
in normal numbers in normal
circumstances, so I'm moving

00:06:38.037 --> 00:06:42.024
the decimal point by one
places into the positive

00:06:42.024 --> 00:06:43.796
direction, so my exponent.

00:06:45.700 --> 00:06:50.991
Will be positive one, but what
does positive one look like in 3

00:06:50.991 --> 00:06:52.212
bits? This compliment?

00:06:52.720 --> 00:06:57.676
So remember I've only got 3
bits. It's a positive number, so

00:06:57.676 --> 00:07:03.458
the 1st digit will be 0 and it's
just positive one, so it's 001.

00:07:04.010 --> 00:07:09.796
So pulling all these things
together and putting it into the

00:07:09.796 --> 00:07:11.374
sign, exponent, exponent,

00:07:11.374 --> 00:07:16.508
mantissa. Format the sign
negative number will be one. The

00:07:16.508 --> 00:07:22.124
exponent is positive one, so it
will be 001 and the months are

00:07:22.124 --> 00:07:26.876
now because the computer knows
that the mantissa is designed to

00:07:26.876 --> 00:07:31.196
be zero point something
something so that zero point. We

00:07:31.196 --> 00:07:36.380
can always forget and I can just
write down the following 4

00:07:36.380 --> 00:07:41.132
digits 1010. So this is
equivalent to minus 1 1/4. So

00:07:41.132 --> 00:07:43.292
this is the 8 bit.

00:07:43.310 --> 00:07:46.094
Floating point notation 4 -- 1

00:07:46.094 --> 00:07:51.742
1/4. Now you might spot here
what happens if I had to put.

00:07:51.742 --> 00:07:56.435
Let's see after normalization, I
end up with five 6 or 7 digits.

00:07:56.435 --> 00:08:01.128
Well with this Ed bits floating
point, we would have to cut off

00:08:01.128 --> 00:08:05.460
at the first 4 bits, so we would
have trunking error. Obviously

00:08:05.460 --> 00:08:09.070
the computer works with bigger
length of digits, therefore that

00:08:09.070 --> 00:08:12.680
trunking error doesn't come up
as much. So in practical

00:08:12.680 --> 00:08:16.290
situation when you have to carry
out the calculations, trunking

00:08:16.290 --> 00:08:17.373
errors do occur.

00:08:17.420 --> 00:08:21.100
In real situations, when the
computer does it, it's very

00:08:21.100 --> 00:08:21.836
very minimized.

00:08:23.060 --> 00:08:25.030
Let's look at another example.

00:08:26.300 --> 00:08:30.240
Minus two and two 8.

00:08:31.420 --> 00:08:33.394
Now when I look at this number.

00:08:34.540 --> 00:08:38.908
Well, I can see here straight
away that I can cancel this one

00:08:38.908 --> 00:08:42.268
down, and this is actually minus
two end of quarter.

00:08:43.040 --> 00:08:45.620
OK, so step one.

00:08:46.500 --> 00:08:52.548
Sign and in this case the sign
will be equal to 1 because I'm

00:08:52.548 --> 00:08:54.708
talking about a negative number.

00:08:54.840 --> 00:08:58.866
Step
2

00:08:58.866 --> 00:09:00.879
convert.

00:09:02.540 --> 00:09:04.520
And find the mantissa.

00:09:05.600 --> 00:09:12.542
So 2 1/4.
Let's convert that

00:09:12.542 --> 00:09:19.066
into binary. Ring in
the place values again, this is

00:09:19.066 --> 00:09:25.534
your radix .124 half or quarter
and an 8. So for two to build

00:09:25.534 --> 00:09:32.464
up the whole part we need 1 zero
and the radix point for 1/4. We

00:09:32.464 --> 00:09:38.470
don't need any half and we use
1/4 so this is our ordinary

00:09:38.470 --> 00:09:44.476
binary which is equal to 2 1/4.
But remember now we need to

00:09:44.476 --> 00:09:49.036
normalize it. And the
normalization process goes as.

00:09:49.990 --> 00:09:53.206
Bring it into the zero point.

00:09:53.930 --> 00:09:57.531
One and the rest of the digit
would follow, so if I normalize

00:09:57.531 --> 00:10:00.855
it, what would means I would
need to move this decimal point

00:10:00.855 --> 00:10:02.794
into the front in here, so I

00:10:02.794 --> 00:10:06.140
would have. The point moved into

00:10:06.140 --> 00:10:12.735
here. And then
what would follow

00:10:12.735 --> 00:10:15.888
now is 0.1001.

00:10:16.290 --> 00:10:20.346
OK, and remember that this part
will be our mantissa, the one

00:10:20.346 --> 00:10:22.036
that follows after the point.

00:10:23.250 --> 00:10:30.498
Then step three is
we need to identify

00:10:30.498 --> 00:10:33.922
or. Express the

00:10:33.922 --> 00:10:38.847
exponent. Now remember the
easiest way to think about the

00:10:38.847 --> 00:10:43.170
exponent is how many places
you would need to move the

00:10:43.170 --> 00:10:47.493
decimal point. If you were the
computer. So you were reading

00:10:47.493 --> 00:10:51.816
this number and the origonal
number was one 0.01, so you

00:10:51.816 --> 00:10:56.532
need to tell the computer via
the exponent that he needs to

00:10:56.532 --> 00:11:01.248
get to this number from this
number. So what would you need

00:11:01.248 --> 00:11:05.964
to do in here? You would need
to move the decimal point.

00:11:07.070 --> 00:11:11.640
Two places into this direction,
and this is the positive

00:11:11.640 --> 00:11:17.124
direction, so the exponent would
have to be positive too, and the

00:11:17.124 --> 00:11:21.237
positive two in three bits 2's
complement. Remember positive

00:11:21.237 --> 00:11:27.178
number is zero and two is 1
zero, so this is our exponent

00:11:27.178 --> 00:11:32.662
now. And Lastly, pull all these
bits together. The sign bit was

00:11:32.662 --> 00:11:39.945
one. The three bits of the
exponent is 010, and the four

00:11:39.945 --> 00:11:42.205
bits of the mantissa.

00:11:42.210 --> 00:11:48.447
Is 1001, which is
this bit here.

00:11:49.190 --> 00:11:55.352
So this is
equal to minus

00:11:55.352 --> 00:11:59.460
two and two
eighths.

00:12:00.980 --> 00:12:08.000
Now look at the last example
and the locks. That last example

00:12:08.000 --> 00:12:15.020
will be minus 116th sign bit
which is. First step is again

00:12:15.020 --> 00:12:19.700
one because it's a negative
number, Step 2.

00:12:20.290 --> 00:12:23.060
Now I need to convert.

00:12:24.090 --> 00:12:26.850
And find the mantissa.

00:12:29.390 --> 00:12:35.198
So place values. I don't have
any whole numbers, so all I'm

00:12:35.198 --> 00:12:40.038
going to do, I'm going to
express the fractional place

00:12:40.038 --> 00:12:42.942
values or half a quarter on

00:12:42.942 --> 00:12:49.485
AIDS. And the 16 and
again because of the way

00:12:49.485 --> 00:12:54.405
I've chosen this number,
this will be 0.0001.

00:12:55.680 --> 00:13:01.389
OK that remember that your
monthly site has to be 0.1

00:13:01.389 --> 00:13:07.617
something. So how can I get from
this number into this format

00:13:07.617 --> 00:13:14.883
now? The only way I can get
to here is by moving the decimal

00:13:14.883 --> 00:13:16.440
.123 places so.

00:13:17.440 --> 00:13:21.864
The exponent, which is
our Step 3.

00:13:24.540 --> 00:13:28.140
I moved my decimal .3 places.

00:13:28.640 --> 00:13:31.780
But have I moved it into
positive or negative direction?

00:13:31.780 --> 00:13:35.548
Let's look at that again. So a
normalized it. But remember that

00:13:35.548 --> 00:13:39.630
mantis and needs to be 4 bits
long, so I need to put.

00:13:41.230 --> 00:13:45.250
Three more zeros in here and
now. The computer comes in and

00:13:45.250 --> 00:13:49.270
you telling the computer that
from this number you want to go

00:13:49.270 --> 00:13:50.945
back to the original number.

00:13:51.620 --> 00:13:56.014
OK, so from here to get back
to the origin number of 0.3

00:13:56.014 --> 00:14:00.070
zeros followed by the one you
have to tell the computer to

00:14:00.070 --> 00:14:01.422
move the radix point.

00:14:02.850 --> 00:14:08.878
Into the opposite direction done
before, so this exponent will be

00:14:08.878 --> 00:14:15.973
negative 3. Now let's see
if we can express negative 3IN.

00:14:16.670 --> 00:14:18.530
3 bits, two complements.

00:14:19.820 --> 00:14:25.309
Remember how we did that is
compliment. If it comes to

00:14:25.309 --> 00:14:30.798
negative numbers, just Express 3
as a 3 bit binary number,

00:14:30.798 --> 00:14:35.289
positive three and positive
three again using the place

00:14:35.289 --> 00:14:40.279
values is 124. Remember, we
don't have anything else in

00:14:40.279 --> 00:14:46.394
here. So it's 011 and to make
it into negative three you need

00:14:46.394 --> 00:14:51.620
to copy the digits until you
copy the one which is in our

00:14:51.620 --> 00:14:56.444
case is the very first step and
then invert everything else swap

00:14:56.444 --> 00:15:01.670
once for zeros and zeros for
once for 01. So this is your

00:15:01.670 --> 00:15:03.278
negative three in here.

00:15:03.790 --> 00:15:05.710
So your exponent.

00:15:07.690 --> 00:15:11.320
He is 101.

00:15:11.990 --> 00:15:17.210
Put the whole thing
together. Sign bit was one,

00:15:17.210 --> 00:15:22.430
the exponent is 101 and a
month is soft.

00:15:24.280 --> 00:15:31.573
Is 1 followed by 3 zero, so
this is the equivalent of minus

00:15:31.573 --> 00:15:33.256
1 / 16.

00:15:35.140 --> 00:15:38.803
Not just to recap, with the
computer would actually do in

00:15:38.803 --> 00:15:42.799
here. The computer would come
and start to read the digits OK

00:15:42.799 --> 00:15:46.129
with this digit. Here you
telling the computer that he

00:15:46.129 --> 00:15:49.792
needs to think about a negative
number. With this digit you

00:15:49.792 --> 00:15:53.122
telling the computer how many
places in what direction he

00:15:53.122 --> 00:15:57.118
needs to move the radix point
that starts from here and then.

00:15:57.118 --> 00:16:01.447
That way it will be able to re
normalize the mantissa. Find out

00:16:01.447 --> 00:16:03.112
what number hides behind this

00:16:03.112 --> 00:16:09.040
binary code. I hope you not have
a clear idea of how to convert

00:16:09.040 --> 00:16:12.540
numbers into 8 bits floating
point notation on the following

00:16:12.540 --> 00:16:15.690
pages you will have some
properties questions and then

00:16:15.690 --> 00:16:17.790
you will find the answers to

00:16:17.790 --> 00:16:21.080
these. So these are the
practice questions.

00:16:27.060 --> 00:16:28.760
And here are the answers.